Courseiva

AWS Certified Machine Learning Specialty MLS-C01 (MLS-C01) — Questions 1–75

1672 questions total · 23pages · All types, answers revealed

Take a mock examExam hub

Page 1 of 23

Page 2
1
Multi-Selecthard

Which THREE factors should be considered when choosing between a parametric and a non-parametric machine learning model?

Select 3 answers
A.Non-parametric models are generally more flexible
B.Parametric models have lower bias than non-parametric models
C.Parametric models train faster than non-parametric models
D.Non-parametric models are less prone to overfitting
E.Parametric models typically require less training data
AnswersA, C, E

Non-parametric models can fit complex patterns.

Why this answer

Non-parametric models, such as k-nearest neighbors or decision trees, do not assume a fixed functional form for the data, allowing them to capture complex, non-linear relationships. This flexibility makes them well-suited for datasets where the underlying distribution is unknown or highly irregular, but it also increases the risk of overfitting if not properly regularized.

Exam trap

AWS often tests the misconception that non-parametric models are less prone to overfitting because they are 'simpler,' when in fact their flexibility makes them more susceptible to overfitting without careful tuning or large datasets.

Full explanation →
2
MCQeasy

A data scientist needs to perform feature scaling for a dataset containing numerical features with different units (e.g., age in years and income in dollars). Which scaling method is most appropriate when the algorithm assumes data is normally distributed?

A.Standardization (Z-score normalization)
B.Log transformation
C.Min-Max scaling
D.Robust scaling (using median and IQR)
AnswerA

Standardization centers data around zero with unit variance, suitable for normality assumptions.

Why this answer

Standardization (Z-score normalization) transforms features to have mean 0 and standard deviation 1, which is suitable for algorithms that assume normally distributed data. Option B (Log transformation) is used to reduce skewness, not for scaling when normality is already assumed. Option C (Min-Max scaling) scales to a fixed range (e.g., 0-1) but does not preserve normal distribution properties.

Option D (Robust scaling) uses median and IQR, which is robust to outliers but does not assume normality.

Full explanation →
3
MCQeasy

A machine learning engineer is exploring a dataset with 500 features and 10,000 samples. To reduce dimensionality for visualization, which technique is most suitable if the goal is to preserve global data structure?

A.t-Distributed Stochastic Neighbor Embedding (t-SNE)
B.Locally Linear Embedding (LLE)
C.Principal Component Analysis (PCA)
D.Uniform Manifold Approximation and Projection (UMAP)
AnswerC

PCA preserves global variance (covariance structure).

Why this answer

PCA is the most suitable technique for preserving the global data structure when reducing dimensionality because it is a linear method that maximizes variance along orthogonal principal components, capturing the overall covariance structure of the 500 features. Unlike nonlinear methods, PCA ensures that the global relationships (e.g., distances between clusters) are retained, making it ideal for visualization of high-dimensional data where the goal is to see broad patterns.

Exam trap

The MLS-C01 exam often tests the misconception that nonlinear methods like t-SNE or UMAP are always better for visualization, but the trap here is that they sacrifice global structure for local detail, making PCA the correct choice when the question explicitly states 'preserve global data structure.'

How to eliminate wrong answers

Option A is wrong because t-SNE is a nonlinear technique that focuses on preserving local neighborhoods and pairwise similarities, often distorting global structure (e.g., cluster sizes and distances) to create visually separable clusters. Option B is wrong because LLE is a nonlinear manifold learning method that preserves local linear relationships between neighbors, but it does not guarantee preservation of global structure and can fail with high-dimensional data (500 features) due to the curse of dimensionality. Option D is wrong because UMAP, while faster than t-SNE, is also a nonlinear technique designed to preserve local and some global structure but prioritizes topological connectivity over global variance, making it less suitable than PCA when the explicit goal is to maintain the overall data covariance and global distances.

Full explanation →
4
Multi-Selecteasy

Which TWO services can be used to serve machine learning models for real-time inference? (Select TWO.)

Select 2 answers
A.Amazon Rekognition
B.Amazon Athena
C.Amazon SageMaker
D.Amazon Comprehend
E.AWS Lambda
AnswersC, E

SageMaker endpoints are designed for real-time inference.

Why this answer

Amazon SageMaker provides real-time endpoints. AWS Lambda can be used with a custom container for inference. Amazon Rekognition is a specialized service for image analysis, not general model serving.

Amazon Comprehend is NLP service. Amazon Athena is query service.

Full explanation →
5
MCQhard

A company uses Amazon EMR with Spark to process data daily. The job reads from S3 and writes to S3. Recently, the job started failing with 'S3AccessDenied' errors. The IAM role used by EMR has not changed. What is the MOST likely cause?

A.The EMR cluster's security group blocks outbound traffic
B.The S3 bucket policy was updated to deny access to the EMR role
C.The S3 bucket was deleted and recreated
D.The EMR service role was rotated
AnswerB

Bucket policies can deny access even if IAM allows.

Why this answer

The IAM role used by EMR has not changed, but the S3 bucket policy can be updated independently to deny access to that specific role. An explicit deny in a bucket policy overrides any allow in the IAM policy, causing the 'S3AccessDenied' error even though the role itself remains unchanged. This is the most likely cause because the job was previously working and only the bucket policy could have been modified without touching the EMR configuration.

Exam trap

The trap here is that candidates assume the IAM role is the only factor in access control, overlooking that S3 bucket policies can be modified independently and can explicitly deny access to a specific role, causing 'AccessDenied' errors even when the role itself is unchanged.

How to eliminate wrong answers

Option A is wrong because security groups control network traffic, not IAM authorization; an outbound block would cause a timeout or connection failure, not an 'S3AccessDenied' error. Option C is wrong because deleting and recreating a bucket with the same name would result in a 'NoSuchBucket' error, not an access denied error, and the bucket name would need to be globally unique. Option D is wrong because rotating the EMR service role would change the role's credentials or ARN, which would break the job entirely (e.g., 'AssumeRole' failures), not produce an 'S3AccessDenied' error; the question states the IAM role used by EMR has not changed.

Full explanation →
6
MCQmedium

A data scientist is working on a regression problem with a dataset that contains outliers. The data scientist is choosing between mean squared error (MSE) and mean absolute error (MAE) as the loss function. Which loss function is more robust to outliers?

A.Both are equally robust to outliers.
B.MSE, because it penalizes large errors more heavily.
C.MAE, because it treats all errors equally.
D.Neither is robust; use Huber loss instead.
AnswerC

MAE is linear in errors, reducing the impact of outliers.

Why this answer

MAE is more robust to outliers because it uses the absolute difference between predicted and actual values, which does not disproportionately penalize large errors. In contrast, MSE squares the errors, causing outliers to have a much larger influence on the loss and model updates. This makes MAE less sensitive to extreme values in regression tasks.

Exam trap

AWS often tests the misconception that a loss function that penalizes errors more heavily is better for robustness, when in fact the opposite is true for outliers.

How to eliminate wrong answers

Option A is wrong because MSE and MAE handle outliers very differently due to their mathematical formulations, so they are not equally robust. Option B is wrong because MSE's heavier penalization of large errors actually makes it less robust to outliers, not more. Option D is wrong because while Huber loss is indeed more robust than both MSE and MAE in some cases, the question asks which of the two given loss functions is more robust, and MAE is the correct choice; Huber loss is not an option in the original comparison.

Full explanation →
7
MCQeasy

A data engineer is building a data pipeline that ingests streaming data from IoT devices. The data must be processed in near real-time and stored in Amazon S3 for further analysis. Which AWS service should be used to capture and process the streaming data before storing it in S3?

A.Use Amazon S3 with S3 Event Notifications to trigger AWS Lambda for processing.
B.Use AWS Glue to perform ETL on the streaming data.
C.Use Amazon Kinesis Data Streams to capture the data and Amazon Kinesis Data Firehose to deliver it to S3.
D.Use Amazon Simple Queue Service (SQS) to buffer the data and then process it with AWS Lambda.
AnswerC

Kinesis Data Streams ingests real-time data and Kinesis Data Firehose delivers it to S3.

Why this answer

Amazon Kinesis Data Streams is designed for real-time data ingestion and can capture streaming data from IoT devices with low latency. Amazon Kinesis Data Firehose then reliably loads that streaming data into Amazon S3, handling buffering, compression, and partitioning automatically. This combination provides the near real-time processing and durable storage required for the pipeline.

Exam trap

The trap here is that candidates often confuse batch ETL services (like AWS Glue) or message queues (like SQS) with purpose-built streaming ingestion services, failing to recognize that Kinesis Data Streams and Firehose are the only AWS-native combination that captures, buffers, and delivers streaming data to S3 in near real-time without custom code.

How to eliminate wrong answers

Option A is wrong because S3 Event Notifications are triggered after data is already stored in S3, not for capturing or processing streaming data before storage; they lack the low-latency ingestion and buffering needed for near real-time streaming. Option B is wrong because AWS Glue is a batch ETL service designed for scheduled or on-demand transformation of data at rest, not for continuous, near real-time capture and processing of streaming data. Option D is wrong because Amazon SQS is a message queue service that does not natively support streaming data ingestion or direct delivery to S3; it would require custom Lambda logic to buffer and write to S3, adding complexity and latency compared to the purpose-built Kinesis Data Firehose integration.

Full explanation →
8
MCQhard

A company is running a data pipeline that uses Amazon EMR with Spark to process 100 TB of data daily. The pipeline must complete within 6 hours. Currently, it takes 8 hours. Which optimization will most likely reduce the runtime?

A.Consolidate small input files into fewer larger files
B.Enable EMR Managed Scaling
C.Increase the memory of each node by using r5 instances
D.Use Spot Instances for all core nodes
AnswerB

Managed Scaling dynamically adds resources to meet deadlines.

Why this answer

EMR Managed Scaling automatically adjusts the number of core and task nodes based on workload, increasing parallelism and reducing runtime for large data pipelines. This is the most effective optimization when the current cluster is under-provisioned for the 100 TB workload. Option A is incorrect because consolidating input files reduces small file overhead but does not address a compute capacity bottleneck.

Option C is incorrect because increasing memory per node (e.g., r5 instances) helps memory-bound tasks, but the primary bottleneck here is likely throughput and parallelism, not memory. Option D is incorrect because Spot Instances can be interrupted, which may cause delays and increase runtime, making them unsuitable for a time-sensitive pipeline.

Full explanation →
9
MCQeasy

A data scientist needs to evaluate a binary classification model's performance. Which metric is most appropriate when the cost of false positives is very high?

A.Accuracy
B.F1 score
C.Recall
D.Precision
AnswerD

Minimizes false positives.

Why this answer

Precision is the most appropriate metric when the cost of false positives is very high because it measures the proportion of positive identifications that were actually correct. In binary classification, precision = TP / (TP + FP), so a high precision means very few false positives occur, directly minimizing the costly error type.

Exam trap

The trap here is that candidates often confuse precision with recall or F1 score, mistakenly thinking that minimizing false positives is best achieved by maximizing recall or a balanced metric, rather than directly optimizing precision.

How to eliminate wrong answers

Option A is wrong because accuracy considers both false positives and false negatives equally, and can be misleading when classes are imbalanced; it does not specifically penalize false positives. Option B is wrong because the F1 score is the harmonic mean of precision and recall, balancing both false positives and false negatives, so it does not prioritize minimizing false positives over false negatives. Option C is wrong because recall measures the proportion of actual positives correctly identified (TP / (TP + FN)), which focuses on avoiding false negatives, not false positives.

Full explanation →
10
MCQmedium

A company uses Amazon SageMaker to train a model. The training job fails with an 'OutOfMemory' error. The training data is stored in S3 and the instance type is ml.m5.xlarge. What is the most efficient way to resolve this issue?

A.Enable managed spot training
B.Reduce the batch size in the training script
C.Increase the number of instances using distributed training
D.Use a larger instance type, such as ml.m5.2xlarge
AnswerD

Larger instance provides more memory.

Why this answer

The 'OutOfMemory' error indicates that the ml.m5.xlarge instance (4 vCPUs, 16 GiB memory) does not have enough RAM to hold the training data and model during processing. Upgrading to ml.m5.2xlarge (8 vCPUs, 32 GiB memory) directly increases available memory, resolving the issue without requiring code changes or architectural modifications. This is the most efficient solution because it requires no script alterations and leverages SageMaker's built-in instance scaling.

Exam trap

The trap here is that candidates often choose 'Reduce the batch size' (Option B) as a quick fix, but the question asks for the 'most efficient' solution—changing instance type requires no code changes and is faster to implement, whereas batch size reduction requires debugging and retesting the training script.

How to eliminate wrong answers

Option A is wrong because managed spot training reduces cost by using spare EC2 capacity but does not increase memory capacity; it can actually cause interruptions that exacerbate resource issues. Option B is wrong because reducing batch size decreases memory usage per step but may not resolve the OOM error if the model itself or the total dataset size exceeds instance memory; it also requires code changes and can slow convergence. Option C is wrong because increasing the number of instances via distributed training (e.g., data parallelism) does not increase the memory of a single instance; each instance still has only 16 GiB, so the OOM error would persist on each worker.

Full explanation →
11
MCQhard

A machine learning engineer is training a neural network on Amazon SageMaker using a custom Docker container. The training job fails with an error: 'CUDA out of memory.' The training instance is an ml.p3.2xlarge with 16 GB GPU memory. The model and data fit into memory when using batch size 32, but the engineer wants to maximize GPU utilization. Which approach should the engineer use to fix the out-of-memory error while maintaining efficient training?

A.Enable mixed precision training
B.Reduce batch size to 1
C.Use a CPU-only instance
D.Implement gradient accumulation with a larger effective batch size
AnswerD

Accumulates gradients over smaller batches to simulate larger batches.

Why this answer

Gradient accumulation allows the engineer to simulate a larger effective batch size by accumulating gradients over multiple forward/backward passes before performing an optimizer step. This keeps the per-step memory footprint low (avoiding CUDA out-of-memory) while maintaining training dynamics similar to a larger batch, thus maximizing GPU utilization without crashing.

Exam trap

The trap here is that candidates may think mixed precision training (Option A) is the direct solution for CUDA out-of-memory, but it only reduces memory per tensor, not the peak memory from batch size; gradient accumulation is the correct technique to handle large effective batches without exceeding GPU memory.

How to eliminate wrong answers

Option A is wrong because mixed precision training reduces memory usage by using float16 for most operations, but it does not directly address the out-of-memory error caused by batch size being too large; it can help but is not the primary fix for this specific error when the goal is to maintain efficient training with a larger effective batch. Option B is wrong because reducing batch size to 1 would drastically reduce GPU utilization and slow convergence, contradicting the goal of maximizing GPU utilization. Option C is wrong because switching to a CPU-only instance would eliminate the GPU entirely, making the training extremely slow and defeating the purpose of using a GPU-accelerated instance like ml.p3.2xlarge.

Full explanation →
12
MCQeasy

A company uses Amazon SageMaker to train a linear regression model. The training data includes a feature 'age' with values ranging from 0 to 100. The model's loss is not converging. What is the MOST likely cause?

A.The features are not normalized.
B.There are outliers in the target variable.
C.The instance type is too small.
D.The learning rate is too high.
AnswerA

Unscaled features cause gradient descent to oscillate.

Why this answer

Normalizing features (e.g., scaling 'age' to a similar range as other features) is critical for linear regression convergence. With 'age' ranging 0-100, if other features are in smaller ranges, the optimizer may struggle. Option B (outliers) can cause issues but is less likely than lack of normalization.

Option C (instance type too small) typically affects training time, not convergence. Option D (learning rate too high) can cause divergence, but normalization is a more common first step.

Full explanation →
13
MCQmedium

A company has deployed a machine learning model on a SageMaker endpoint that serves predictions to a web application. The model uses a custom inference container that loads the model artifacts from an ECR repository. After updating the model with new training data, the data scientist creates a new model and updates the endpoint. However, some users report that they still get predictions from the old model. The data scientist confirms that the endpoint configuration points to the new model. What is the most likely cause?

A.The new model artifacts are not correctly uploaded to S3
B.The endpoint is behind a load balancer that is not updated
C.The inference container is cached and not pulling the new image
D.DNS caching on the client side is resolving to the old endpoint IP address
AnswerD

Correct. DNS caching at the client side can cause the endpoint's DNS name to resolve to an old IP address, particularly if the endpoint's underlying instance IPs have not changed, leading some users to still hit the old model.

Why this answer

DNS caching on the client side can cause the endpoint's DNS name to resolve to an old IP address, especially if the endpoint's underlying instances have not changed. This explains why some users still receive predictions from the old model even though the endpoint configuration points to the new model. Option A is incorrect because incorrect model artifacts would affect all users uniformly.

Option B is incorrect because SageMaker endpoints do not have load balancers in the traditional sense; the endpoint itself routes traffic to the instances. Option C is incorrect because SageMaker automatically handles container image updates when a new model is deployed to an endpoint.

Full explanation →
14
MCQmedium

A data scientist is using Amazon SageMaker to train a model with a custom Docker container. The training script reads data from an S3 bucket and writes the model artifact to an S3 bucket. The training job fails with a 'NoSuchKey' error. What is the MOST likely cause?

A.The training script is not compatible with the Docker image.
B.The training data path specified in the input data channel is incorrect.
C.The Docker image is not available in Amazon ECR.
D.The SageMaker execution role does not have s3:GetObject permission.
AnswerB

NoSuchKey means the S3 key does not exist.

Why this answer

The 'NoSuchKey' error indicates that the specified key (file path) does not exist within the S3 bucket, which occurs when the training data path provided in the input data channel is incorrect. Option A is wrong because, while a script compatibility issue could cause errors, it would not manifest as a 'NoSuchKey' S3 error; it would likely result in a failure during container execution. Option C is wrong because if the Docker image were not available in Amazon ECR, SageMaker would fail with an 'ImageNotFoundException' or similar ECR-related error, not a 'NoSuchKey' S3 error.

Option D is wrong because insufficient S3 permissions (e.g., missing s3:GetObject) would result in an 'AccessDenied' error, not 'NoSuchKey'.

Full explanation →
15
MCQhard

A machine learning team is working with a dataset containing high-dimensional sparse features, such as text data represented as bag-of-words. The team wants to reduce dimensionality while preserving the structure of the sparse matrix. Which technique is most appropriate for this scenario?

A.t-distributed Stochastic Neighbor Embedding (t-SNE).
B.Truncated Singular Value Decomposition (SVD).
C.Linear Discriminant Analysis (LDA).
D.Principal Component Analysis (PCA) using the covariance matrix.
AnswerB

Truncated SVD works efficiently on sparse matrices.

Why this answer

Truncated SVD (e.g., using sklearn's TruncatedSVD) is specifically designed for sparse matrices and efficiently reduces dimensionality while preserving the structure of the sparse matrix. Option A (t-SNE) is primarily for visualization and does not preserve the global structure well for dimensionality reduction. Option C (LDA) is a supervised method that requires labels and is not suitable for unsupervised dimensionality reduction.

Option D (PCA using covariance matrix) requires a dense matrix and is computationally expensive for high-dimensional sparse data.

Full explanation →
16
MCQeasy

A company wants to use Amazon SageMaker to train a deep learning model using a custom TensorFlow script. The data is stored in an S3 bucket. Which SageMaker API operation should be used to launch the training job?

A.CreateHyperParameterTuningJob
B.CreateEndpoint
C.CreateTransformJob
D.CreateTrainingJob
AnswerD

CreateTrainingJob starts a training job.

Why this answer

The correct API operation to launch a training job in Amazon SageMaker is CreateTrainingJob. This operation specifies the training algorithm (or custom script), resource configuration (instance type and count), input data configuration (pointing to S3), and output location for the model artifacts. It directly initiates the training process on SageMaker-managed infrastructure.

Exam trap

The MLS-C01 exam often tests the distinction between training, tuning, inference, and batch transform operations, so candidates mistakenly choose CreateHyperParameterTuningJob when the question only asks for a single training job, or choose CreateEndpoint when they confuse training with deployment.

How to eliminate wrong answers

Option A is wrong because CreateHyperParameterTuningJob is used to launch a hyperparameter tuning job, which runs multiple training jobs with different hyperparameter combinations to find the best model; it is not the direct API for a single training job. Option B is wrong because CreateEndpoint is used to deploy a trained model as a real-time inference endpoint, not to start training. Option C is wrong because CreateTransformJob is used for batch inference on an existing trained model, not for training a new model.

Full explanation →
17
MCQhard

During exploratory data analysis, a data scientist discovers that a feature has a variance of 0.01, while other features have variances around 1.0. Which action should be taken?

A.Scale the feature to have unit variance.
B.Apply a log transformation to the feature.
C.Impute missing values in the feature.
D.Consider removing the feature or applying variance threshold.
AnswerD

Near-zero variance features are often uninformative.

Why this answer

A feature with near-zero variance (0.01) compared to others (~1.0) likely has little predictive power and can cause numerical instability in models. Variance thresholding is a standard preprocessing step to remove low-variance features. Option A is wrong: scaling to unit variance does not address the fundamental issue of low information content; it merely changes the scale.

Option B is wrong: log transformation changes distribution shape but does not meaningfully increase variance; the variance will remain low. Option C is wrong: imputation is for missing values, which is unrelated to low variance.

Full explanation →
18
Multi-Selecthard

A machine learning engineer is designing an automated ML pipeline for training and deploying models. The pipeline must include data validation, model training, hyperparameter tuning, and model deployment. The engineer wants to use AWS services that integrate well and provide version control. Which THREE services should be combined to achieve this? (Choose THREE.)

Select 3 answers
A.AWS Glue
B.AWS Step Functions
C.AWS CodePipeline
D.Amazon EMR
E.Amazon SageMaker
AnswersB, C, E

Correct: Orchestrates the ML pipeline steps.

Why this answer

AWS Step Functions orchestrates the ML pipeline by coordinating data validation, model training, hyperparameter tuning, and deployment as a state machine. It integrates natively with Amazon SageMaker for training and tuning jobs, and with AWS CodePipeline for CI/CD, enabling version control and automated transitions between pipeline stages.

Exam trap

The MLS-C01 exam often tests the distinction between orchestration services (Step Functions) and data processing services (Glue, EMR), leading candidates to mistakenly choose Glue for pipeline orchestration because of its 'data validation' capability, ignoring its lack of ML-specific deployment and tuning features.

Full explanation →
19
MCQmedium

Refer to the exhibit. A data scientist is using an IAM role with this policy to run a SageMaker processing job that reads data from S3. The job fails with an access error. What is the most likely cause?

A.The policy does not allow sagemaker:CreateProcessingJob
B.The policy does not allow s3:PutObject
C.The policy does not allow s3:ListBucket
D.The policy does not allow s3:GetObject
AnswerC

ListBucket is required to list objects in the bucket.

Why this answer

The processing job needs both s3:GetObject and s3:ListBucket to read objects. The policy lacks s3:ListBucket. Option A is wrong because sagemaker:CreateProcessingJob is allowed.

Option B is wrong because s3:PutObject is not needed for reading. Option D is wrong because the policy allows s3:GetObject.

Full explanation →
20
MCQeasy

A data scientist is using Amazon SageMaker to train a model. The training job fails with an error 'Insufficient instance capacity'. Which action should the scientist take to resolve this?

A.Check IAM permissions
B.Retry with a different instance type or Availability Zone
C.Increase the number of instances
D.Reduce the training data size
AnswerB

Different instances/AZs may have available capacity.

Why this answer

Retrying with a different instance type or in a different Availability Zone often resolves capacity issues. Option A is incorrect because the error is about capacity, not IAM permissions. Option C is incorrect because increasing the number of instances may exacerbate the capacity issue.

Option D is incorrect because reducing the training data size does not affect instance capacity.

Full explanation →
21
MCQmedium

A data scientist ran an XGBoost training job on Amazon SageMaker using a CSV dataset. The training job failed with the error shown. What is the most likely cause of this failure?

A.The dataset includes a header row
B.One of the rows in the CSV has an extra column
C.The delimiter used in the CSV is not a comma
D.The dataset contains missing values
AnswerB

The error message indicates that line 1 has 11 fields instead of the expected 10, meaning an extra column in that row.

Why this answer

XGBoost on SageMaker expects the CSV input to have a consistent number of columns across all rows. If one row contains an extra column, the parser will fail because it cannot map the additional field to the feature schema defined by the training job. This mismatch causes the 'Error: Number of columns does not match' or similar parsing error.

Exam trap

The trap here is that candidates often confuse a column count mismatch with missing values or delimiter issues, but the error message specifically points to inconsistent row lengths, not data quality or formatting problems.

How to eliminate wrong answers

Option A is wrong because XGBoost can handle a header row if the `csv_reader` parameter is set to skip it, and a header alone does not cause a column count mismatch error. Option C is wrong because SageMaker's XGBoost implementation defaults to comma delimiter, but if a different delimiter is used, the error would be about unrecognized delimiter or parsing, not a column count mismatch. Option D is wrong because XGBoost natively handles missing values (e.g., by treating them as NaN or using the `missing` parameter), and missing values do not cause a column count error.

Full explanation →
22
MCQmedium

A company uses Amazon SageMaker to deploy a model for real-time inference. The endpoint uses an ml.m5.large instance with automatic scaling based on CPU utilization. The team notices that during traffic spikes, the endpoint returns 5xx errors. What should the team do to improve the endpoint's availability?

A.Increase the instance type to ml.c5.2xlarge.
B.Reduce the scaling cooldown period.
C.Place an Application Load Balancer in front of the endpoint.
D.Use Amazon API Gateway to throttle requests.
AnswerA

Larger instance type provides more capacity.

Why this answer

Upgrading the instance type from ml.m5.large to ml.c5.2xlarge provides more CPU and memory resources, which directly addresses the root cause of 5xx errors during traffic spikes — insufficient compute capacity to handle the request load. Automatic scaling based on CPU utilization may not react quickly enough to sudden spikes, leading to request queuing and timeouts that manifest as 5xx errors. A larger instance type increases the baseline throughput, reducing the likelihood of resource exhaustion before scaling can take effect.

Exam trap

The trap here is that candidates often confuse auto-scaling configuration (cooldown periods, thresholds) with raw capacity planning, assuming that tuning scaling parameters alone can handle sudden spikes, when in fact the instance must have enough headroom to survive the scaling latency.

How to eliminate wrong answers

Option B is wrong because reducing the scaling cooldown period would cause the auto-scaling policy to react more aggressively to short-lived CPU spikes, potentially leading to thrashing and unnecessary instance provisioning, but it does not address the immediate capacity shortfall during the spike itself — the endpoint still needs enough raw compute to handle the burst. Option C is wrong because placing an Application Load Balancer (ALB) in front of the SageMaker endpoint is not supported — SageMaker endpoints use a built-in load balancer (AWS-managed) and cannot have an external ALB inserted; ALBs distribute traffic to targets like EC2 instances, not SageMaker hosted endpoints. Option D is wrong because using Amazon API Gateway to throttle requests would intentionally drop or delay requests during spikes, which would not improve availability but instead cause client-side errors (429 Too Many Requests) and degrade user experience, whereas the goal is to serve all requests successfully.

Full explanation →
23
MCQeasy

A machine learning engineer is working on a regression problem to predict house prices. The dataset contains 500,000 rows and 20 features, including 'sqft_living', 'bedrooms', 'bathrooms', 'floors', 'waterfront', 'view', 'condition', 'grade', 'yr_built', 'zipcode', and 'lat'. After performing exploratory data analysis, the engineer notices that the 'sqft_living' feature has a right-skewed distribution with a long tail. The 'zipcode' feature is categorical with 70 unique values. The 'lat' feature is continuous. The engineer wants to prepare the data for a linear regression model. Which action should the engineer take to improve model performance?

A.Remove the 'sqft_living' feature because it violates the normality assumption.
B.Apply a log transformation to the 'sqft_living' feature.
C.One-hot encode the 'zipcode' feature to capture location effects.
D.Apply standard scaling (z-score) to the 'sqft_living' feature.
AnswerB

Log transformation reduces right skewness, making the distribution more symmetric.

Why this answer

Linear regression assumes that features are approximately normally distributed, and a right-skewed distribution like 'sqft_living' can violate this assumption, leading to poor model performance. Applying a log transformation compresses the long tail, making the distribution more symmetric and helping the model learn a linear relationship between the feature and the target. This is a standard preprocessing step for skewed features in regression tasks.

Exam trap

The MLS-C01 exam often tests the misconception that standard scaling (z-score) can fix skewness, when in reality it only normalizes the mean and variance without altering the shape of the distribution.

How to eliminate wrong answers

Option A is wrong because removing the 'sqft_living' feature outright discards valuable information; linear regression does not require strict normality of features, only that residuals are normally distributed, and skewness can be addressed via transformation. Option C is wrong because one-hot encoding 'zipcode' with 70 unique values would create 69 dummy variables, which is acceptable but not the most impactful action for improving model performance given the stated issue of skewness in 'sqft_living'. Option D is wrong because standard scaling (z-score) only centers and scales the data, which does not address right skewness; it would preserve the long tail and fail to make the distribution more normal.

Full explanation →
24
MCQhard

A company uses a linear regression model to predict house prices. The model's R-squared is 0.95 on the training set but 0.60 on the test set. Which of the following is the most likely cause?

A.Overfitting
B.Underfitting
C.Multicollinearity among features
D.Data leakage
AnswerA

Correct. A high training R-squared with a much lower test R-squared is a classic sign of overfitting, where the model learns noise in the training data but fails to generalize.

Why this answer

Such a large gap between training and test R-squared indicates overfitting. Option B is wrong because if the model were underfitting, both would be low. Option C is wrong because multicollinearity affects both training and test sets similarly and does not typically cause such a gap.

Option D is wrong because data leakage would likely cause high test R-squared, not a significant drop.

Full explanation →
25
MCQhard

A team is training a deep learning model on Amazon SageMaker using a large dataset stored in S3. The training job is taking a long time, and the team suspects that data loading is the bottleneck. The dataset consists of many small files (average size 10KB). Which change would most effectively reduce the I/O bottleneck?

A.Combine the small files into larger files (e.g., TFRecord format)
B.Use SageMaker Pipe mode instead of File mode
C.Increase the number of training instances
D.Use a P3 instance type for better GPU performance
AnswerA

Larger files reduce the number of S3 API calls and improve throughput.

Why this answer

Combining small files into larger ones (e.g., TFRecord or Parquet) reduces the number of S3 GET requests and improves throughput. Using Pipe mode reads data sequentially, which is less efficient for random access. Increasing instance count or using P3 instances addresses compute, not I/O.

Amazon EFS is not recommended for training jobs due to higher latency.

Full explanation →
26
Multi-Selecthard

Which THREE actions can help reduce the inference latency of a SageMaker endpoint? (Choose three.)

Select 3 answers
A.Use a larger instance type with more CPU/GPU
B.Enable SageMaker Batch Transform to process predictions offline
C.Enable data compression to reduce payload size
D.Use a multi-model endpoint to share instances across models
E.Increase the number of instances in the endpoint
AnswersA, B, C

More compute power reduces per-request latency.

Why this answer

Using a larger instance type with more CPU or GPU resources directly increases the computational capacity available for inference. This reduces the time required to process each prediction request, thereby lowering inference latency. SageMaker endpoints scale horizontally and vertically, and vertical scaling (larger instances) is a straightforward way to improve per-request performance.

Exam trap

The MLS-C01 exam often tests the distinction between improving throughput (horizontal scaling) versus reducing latency (vertical scaling or optimization), and candidates mistakenly assume that adding more instances will speed up individual requests.

Full explanation →
27
MCQmedium

A financial services company is building a model to detect fraudulent credit card transactions. The dataset contains 1 million transactions, with only 0.1% labeled as fraud. The data scientist trains a logistic regression model on the raw dataset and obtains the following results on a held-out test set: accuracy = 99.8%, precision = 50%, recall = 60%, F1 = 0.545. The business requirement is to maximize recall while keeping precision above 80%. Which course of action should the data scientist take to improve the model?

A.Use random undersampling of the majority class to balance the dataset
B.Collect more historical transaction data and retrain the model
C.Train the model with class weights inversely proportional to class frequencies
D.Apply L2 regularization with a higher penalty to reduce overfitting
AnswerC

Class weights help the model focus on the minority class, often improving precision and recall.

Why this answer

Assigning class weights inversely proportional to class frequencies penalizes misclassifications of the minority class (fraud) more heavily during training. This directly addresses the severe class imbalance (0.1% fraud) by forcing the logistic regression model to learn decision boundaries that improve recall, while the weight ratio can be tuned to maintain precision above 80%. Unlike naive resampling, this approach preserves the original data distribution and avoids information loss.

Exam trap

The MLS-C01 exam often tests the misconception that resampling (undersampling or oversampling) is always the best first step for imbalance, when in fact cost-sensitive learning via class weights is often more effective and stable for linear models like logistic regression.

How to eliminate wrong answers

Option A is wrong because random undersampling of the majority class discards a large number of legitimate transactions, which can lead to loss of valuable patterns and increased variance, often degrading precision rather than maintaining it above 80%. Option B is wrong because simply collecting more historical data does not change the underlying class imbalance ratio; without addressing the imbalance, the model will still be biased toward the majority class and fail to improve recall. Option D is wrong because L2 regularization with a higher penalty reduces overfitting by shrinking coefficients, but it does not target class imbalance; it may actually worsen recall by further suppressing the already weak signal from the minority class.

Full explanation →
28
MCQmedium

Refer to the exhibit. A data scientist is using AWS Glue ETL jobs to process data from a source database. The job logs show repeated timeout errors. Which EDA step should the scientist perform to diagnose the issue?

A.Test network connectivity from the Glue job to the source database using telnet.
B.Check the source database table sizes and row counts over time.
C.Switch the Glue ETL job type from Spark to Python shell to reduce overhead.
D.Increase the Glue job timeout to 600 seconds and rerun.
AnswerB

Identifies if data volume growth causes timeouts.

Why this answer

The timeout errors indicate that the Glue ETL job is exceeding its configured timeout. To diagnose the root cause, the data scientist should check the source database table sizes and row counts over time (option B). This helps determine if the data volume has increased, causing longer processing times.

Option A (testing network connectivity) addresses network issues but not processing delays. Option C (switching job type) may change performance but does not diagnose the cause. Option D (increasing timeout) is a temporary workaround, not a diagnostic step.

Full explanation →
29
MCQmedium

You are deploying a PyTorch model to a SageMaker endpoint. The model is large (5 GB) and the endpoint is using an ml.c5.2xlarge instance. Inference latency is higher than required. Which change would most effectively reduce latency?

A.Reduce the batch size in the inference code
B.Decrease the number of model server workers
C.Enable SageMaker Elastic Inference
D.Use a GPU instance type such as ml.p3.2xlarge
AnswerD

GPU accelerates matrix operations in PyTorch.

Why this answer

The primary bottleneck for a large PyTorch model (5 GB) on a CPU instance (ml.c5.2xlarge) is the lack of GPU acceleration for matrix operations and tensor computations. Switching to a GPU instance like ml.p3.2xlarge (with NVIDIA V100 GPUs) offloads the heavy parallel computation to the GPU, drastically reducing per-inference latency for deep learning models, especially those with large parameter counts.

Exam trap

The trap here is that candidates often choose Elastic Inference (Option C) thinking it provides GPU-like acceleration at lower cost, but they overlook the model size limitation (max ~2 GB) and the added network latency, making it unsuitable for large models like a 5 GB PyTorch model.

How to eliminate wrong answers

Option A is wrong because reducing batch size decreases throughput and may increase per-request overhead, but it does not address the fundamental CPU compute bottleneck for a 5 GB model; latency per inference often remains high due to sequential CPU processing. Option B is wrong because decreasing the number of model server workers reduces concurrency and can increase queueing delay, but it does not accelerate the actual inference computation on the CPU. Option C is wrong because SageMaker Elastic Inference attaches a small, shared GPU accelerator for specific operations (e.g., matrix multiply), but it is not designed for models larger than ~2 GB and introduces network latency for data transfer between the CPU instance and the EI accelerator, making it ineffective for a 5 GB model.

Full explanation →
30
MCQmedium

A data scientist is analyzing a dataset of customer reviews for a retail company. The dataset contains text reviews, star ratings (1-5), and customer metadata. The scientist wants to perform sentiment analysis to classify reviews as positive or negative. During EDA, the scientist uses Amazon SageMaker Data Wrangler to visualize the distribution of star ratings and notices that 90% of reviews are 4 or 5 stars, while only 2% are 1 star. The scientist is concerned about class imbalance. Which approach should the scientist take to address the imbalance before modeling?

A.Downsample the majority class to create a balanced dataset.
B.Use random oversampling of the minority class to balance the dataset.
C.Use accuracy as the evaluation metric since it is simple.
D.Use the F1-score as the evaluation metric to account for imbalance.
AnswerD

F1-score balances precision and recall, appropriate for imbalanced classes.

Why this answer

The F1-score balances precision and recall, making it a suitable metric for imbalanced datasets where accuracy would be misleading due to the majority class dominating. Option A (downsampling) is not ideal because it discards potentially useful data from the majority class. Option B (random oversampling) can lead to overfitting by duplicating minority class instances, and is not a guaranteed solution.

Option C is incorrect because accuracy is not reliable when classes are imbalanced, as high accuracy can be achieved by simply predicting the majority class.

Full explanation →
31
MCQmedium

A company uses SageMaker built-in BlazingText algorithm for text classification. The model performance is poor on the validation set. The data consists of short documents (average 50 words). Which hyperparameter tuning strategy is most likely to improve performance?

A.Increase bucket size from 0 to 1000000
B.Increase vector dimension from 100 to 300
C.Increase minCount from 1 to 5
D.Decrease window size from 5 to 2
AnswerD

Smaller window size captures local context better for short documents.

Why this answer

BlazingText's default window size of 5 may be too large for short documents (average 50 words), causing the model to learn overly broad context that dilutes local semantic patterns. Decreasing the window size to 2 forces the model to focus on tighter word co-occurrences, which is more effective for short text classification where local n-gram signals are critical.

Exam trap

The trap here is that candidates often assume increasing model capacity (e.g., vector dimension) or filtering rare words (minCount) always helps, but for short documents, the hyperparameter controlling context granularity (window size) is the most impactful lever.

How to eliminate wrong answers

Option A is wrong because bucket size controls subword n-gram hashing for out-of-vocabulary words, not classification performance on short documents; increasing it from 0 to 1,000,000 would add computational overhead without addressing the core issue of context window. Option B is wrong because increasing vector dimension from 100 to 300 risks overfitting on small short-text datasets and does not fix the problem of overly broad context capture. Option C is wrong because increasing minCount from 1 to 5 discards rare but potentially discriminative words in short documents, further reducing signal in an already sparse dataset.

Full explanation →
32
MCQhard

A data scientist is using Amazon SageMaker Data Wrangler to perform exploratory data analysis on a dataset. The dataset contains a feature 'age' with values ranging from 0 to 120. The data scientist wants to detect outliers. Which built-in transform in Data Wrangler is most appropriate for this task?

A.Handle Outliers
B.Scale and Normalize
C.Handle Missing
D.Encode Categorical
AnswerA

This transform includes outlier detection methods such as IQR and z-score.

Why this answer

The 'Handle Outliers' transform in Amazon SageMaker Data Wrangler provides built-in methods such as IQR (Interquartile Range) and z-score to detect and handle outliers in numeric features like 'age'. Option B (Scale and Normalize) is used to rescale features but does not detect outliers. Option C (Handle Missing) deals with missing values, not outliers.

Option D (Encode Categorical) is for converting categorical variables to numerical, not for outlier detection.

Full explanation →
33
MCQhard

A data scientist is performing exploratory data analysis on text data. They want to identify the most common terms and their frequencies. Which approach should they use?

A.Perform sentiment analysis on the text.
B.Apply Latent Dirichlet Allocation (LDA) to extract topics.
C.Create a bag-of-words matrix and compute term frequencies.
D.Use word2vec to generate word embeddings.
AnswerC

Bag-of-words directly provides term counts.

Why this answer

A bag-of-words matrix counts the frequency of each term in the text, directly providing the most common terms and their frequencies. Option A is incorrect because sentiment analysis determines the emotional tone of text, not term frequencies. Option B is incorrect because Latent Dirichlet Allocation (LDA) is a topic modeling technique that assigns topics to documents, not term frequencies.

Option D is incorrect because word2vec generates dense vector embeddings that capture semantic relationships, not raw term frequencies.

Full explanation →
34
MCQmedium

A data scientist is working with a dataset containing customer transactions. The dataset has a column named 'transaction_date' with timestamp values. The scientist wants to create new features such as day of week, hour, and whether the transaction occurred on a weekend. Which AWS service provides built-in feature engineering capabilities for datetime columns?

A.Amazon SageMaker Data Wrangler
B.Amazon Athena
C.AWS Glue ETL
D.Amazon EMR
AnswerA

Amazon SageMaker Data Wrangler includes built-in transformations for datetime features.

Why this answer

Amazon SageMaker Data Wrangler includes built-in transformations for datetime features like extracting day, month, hour, etc. Option B (Amazon Athena) can extract parts but not as a feature engineering step. Option C (AWS Glue ETL) requires custom code.

Option D (Amazon EMR) requires more manual effort.

Full explanation →
35
MCQmedium

Refer to the exhibit. A data scientist is unable to run an Amazon Athena query on data in `my-bucket`. The IAM policy shown is attached to the user. What is the most likely reason for the failure?

A.The ListBucket action is not granted.
B.Athena needs s3:PutObject permission to write results.
C.The data is encrypted with SSE-C.
D.The bucket does not exist.
AnswerB

Athena writes output to S3.

Why this answer

Athena requires `s3:PutObject` permission to write query results to an output location. Without this permission, the query fails. Option A is not the issue because the bucket exists and is accessible; Option C is not necessarily relevant; Option D is incorrect because the bucket does exist (otherwise the user couldn't even attempt the query).

Full explanation →
36
MCQmedium

A data scientist is exploring a dataset with many missing values. They want to understand the pattern of missingness before deciding on imputation. Which approach is most appropriate?

A.Compute the correlation matrix of the features with missing values.
B.Drop all rows with any missing values.
C.Impute all missing values with the mean of each column.
D.Visualize the missingness using a heatmap or bar chart.
AnswerD

Visualization helps identify patterns like monotonic or random missingness.

Why this answer

Visualizing missingness with a heatmap or bar chart (using libraries like missingno) reveals patterns such as MCAR, MAR, or MNAR. Option A (correlation matrix) does not directly show missingness patterns. Option B (dropping rows) may remove valuable data and assumes MCAR.

Option C (mean imputation) also assumes MCAR and can bias results if missingness is not random.

Full explanation →
37
MCQeasy

A data scientist is training a binary classification model on a dataset with 10,000 features. The model overfits severely. Which technique is MOST appropriate to reduce overfitting?

A.Apply L1 regularization (Lasso)
B.Use early stopping during training
C.Use PCA to reduce dimensionality
D.Increase the max depth of the model
AnswerA

L1 regularization penalizes the absolute size of coefficients, driving some to zero and reducing overfitting.

Why this answer

With 10,000 features, L1 regularization (Lasso) is the most appropriate technique because it performs feature selection by shrinking less important feature coefficients to exactly zero. This directly addresses the high-dimensional overfitting by reducing model complexity and removing noise features, which is more effective than other methods for such a large feature space.

Exam trap

The MLS-C01 exam often tests the misconception that dimensionality reduction (PCA) is always the best solution for high-dimensional overfitting, but L1 regularization is more direct because it performs feature selection within the model itself, preserving interpretability and avoiding information loss from linear projections.

How to eliminate wrong answers

Option B is wrong because early stopping prevents overfitting by halting training when validation loss stops improving, but it does not address the root cause of high dimensionality (10,000 features) and is less effective than regularization for feature selection. Option C is wrong because PCA reduces dimensionality via linear projection, but it loses interpretability and may discard important features; it is a preprocessing step, not a regularization technique, and can still overfit if the model is complex. Option D is wrong because increasing max depth increases model complexity, which worsens overfitting rather than reducing it.

Full explanation →
38
MCQeasy

A data scientist is training a binary classification model using Amazon SageMaker's XGBoost. The dataset is highly imbalanced (99% negative class, 1% positive class). The data scientist wants to maximize the F1-score. Which parameter adjustment is most appropriate?

A.Set max_depth to 10
B.Set eta to 0.01
C.Set subsample to 0.5
D.Set scale_pos_weight to 99
AnswerD

scale_pos_weight adjusts the balance of positive and negative weights; a value of 99 (ratio of negatives to positives) helps the model focus on the minority class.

Why this answer

Setting scale_pos_weight to 99 is the most appropriate adjustment because it directly addresses class imbalance by assigning a higher weight to the minority (positive) class during training. In XGBoost, scale_pos_weight controls the balance of positive and negative weights, typically set as sum(negative instances) / sum(positive instances), which here is 99/1 = 99. This forces the model to penalize misclassifications of the positive class more heavily, thereby improving recall and F1-score.

Exam trap

The trap here is that candidates may confuse hyperparameters that control model complexity (max_depth, subsample) or learning rate (eta) with those that directly handle class imbalance, missing that scale_pos_weight is the specific XGBoost parameter designed for this purpose.

How to eliminate wrong answers

Option A is wrong because increasing max_depth to 10 can lead to overfitting, especially on imbalanced data, and does not directly address class imbalance or optimize F1-score. Option B is wrong because setting eta (learning rate) to 0.01 reduces the step size for updates, which can slow convergence but does not specifically handle class imbalance or improve F1-score. Option C is wrong because setting subsample to 0.5 randomly samples 50% of the training data per tree, which may reduce overfitting but does not target the imbalance between positive and negative classes.

Full explanation →
39
MCQeasy

Refer to the exhibit. A data scientist is evaluating a binary classification model for spam detection. The exhibit shows a single prediction instance. What is the model's prediction for this instance?

A.Ham
B.0.95
C.The model is unsure because probability is not 1.0
D.Spam
AnswerD

The 'predicted_label' is 'spam'.

Why this answer

The model's prediction is 'Spam' because the prediction instance shows a probability of 0.95 for the 'Spam' class, which exceeds the typical decision threshold of 0.5 used in binary classification. Since the probability for 'Spam' is higher than for 'Ham' (0.05), the model assigns the instance to the class with the highest probability, which is Spam.

Exam trap

The MLS-C01 exam often tests the distinction between a model's probability output and its final class prediction, leading candidates to mistakenly select the probability value (0.95) as the prediction instead of the class label (Spam).

How to eliminate wrong answers

Option A is wrong because 'Ham' is the class with the lower probability (0.05), and the model predicts the class with the highest probability, not the lowest. Option B is wrong because 0.95 is the probability score for the Spam class, not the final class prediction; the model outputs a class label (Spam or Ham), not a probability value as the prediction. Option C is wrong because the model is not 'unsure' — in binary classification, a probability of 0.95 indicates high confidence for the Spam class, and the model always makes a deterministic prediction based on the decision threshold (typically 0.5), regardless of whether the probability is 1.0.

Full explanation →
40
MCQmedium

A machine learning engineer is deploying a sentiment analysis model using Amazon SageMaker. The model is a BERT-based transformer that takes up to 512 tokens. The engineer notices that inference latency is high (over 500 ms per request) on a single ml.c5.xlarge instance. The application requires latency under 100 ms. The model has already been optimized using half-precision (FP16). Which action should the engineer take to reduce latency?

A.Use a GPU instance such as ml.g4dn.xlarge
B.Reduce the maximum sequence length to 128
C.Increase the batch size for inference requests
D.Use SageMaker Neo to compile the model for the target instance
AnswerA

GPUs accelerate transformer inference significantly.

Why this answer

(Use a GPU instance such as ml.g4dn.xlarge) provides the parallel processing power needed to accelerate transformer inference, significantly reducing latency for BERT-based models even after FP16 optimization. Option B (Reduce max sequence length to 128) would lower computational cost and latency but at the risk of truncating input and degrading accuracy, so it is not the primary recommended action. Option C (Increase batch size) improves throughput for multiple requests but does not reduce the latency of a single request.

Option D (SageMaker Neo) optimizes the model for the target instance, but with an already optimized model (FP16) and strict latency requirement under 100 ms, switching to a GPU instance is more effective.

Full explanation →
41
MCQmedium

A company is building a text classification model to categorize customer support tickets. The dataset is highly imbalanced with 95% of tickets belonging to 'General Inquiry' and 5% to 'Complaint'. The data scientist is using a random forest classifier. Which metric is most appropriate for evaluating model performance on the minority class?

A.Accuracy
B.F1-score for 'Complaint'
C.ROC AUC
D.Precision for 'Complaint'
AnswerB

F1-score balances precision and recall, making it suitable for imbalanced classification.

Why this answer

In a highly imbalanced dataset (95% General Inquiry, 5% Complaint), accuracy is misleading because a model that predicts 'General Inquiry' for every ticket would achieve 95% accuracy but completely fail on the minority class. The F1-score for 'Complaint' is the harmonic mean of precision and recall, providing a balanced evaluation of the model's ability to correctly identify complaints without being skewed by the majority class. For a random forest classifier, this metric directly addresses the minority class performance, which is the primary concern.

Exam trap

The trap here is that candidates often choose accuracy due to its simplicity, failing to recognize that on imbalanced datasets it is a deceptive metric, or they select ROC AUC because it is commonly used for binary classification, but it does not isolate minority class performance as effectively as the F1-score.

How to eliminate wrong answers

Option A is wrong because accuracy is inappropriate for imbalanced datasets; a naive model predicting only the majority class would achieve 95% accuracy, masking poor minority class performance. Option C is wrong because ROC AUC measures the trade-off between true positive rate and false positive rate across all thresholds, which can be overly optimistic on imbalanced data and does not directly reflect precision or recall for the minority class. Option D is wrong because precision for 'Complaint' alone ignores recall; a model could achieve high precision by making very few positive predictions (e.g., only when extremely confident), but miss most actual complaints, which is unacceptable for detecting customer complaints.

Full explanation →
42
MCQmedium

A data scientist is training a binary classifier using logistic regression on a dataset that is highly imbalanced (95% negative class, 5% positive class). The model achieves 95% accuracy but only predicts the negative class. Which metric should the scientist use to evaluate the model's performance on the positive class?

A.Recall
B.Precision
C.F1 Score
D.Accuracy
AnswerA

Correct. Recall measures the proportion of actual positives correctly identified. Since the model never predicts positive, recall is 0, which accurately reflects the model's inability to detect positive instances.

Why this answer

The model only predicts the negative class, so recall (true positives / (true positives + false negatives)) is 0, directly revealing the failure to detect positives. Precision is undefined (or 0/0) because there are no positive predictions. Accuracy is misleading (95% but useless).

Therefore, recall is the most informative metric for evaluating performance on the positive class.

Exam trap

Candidates may think precision is better because it focuses on correctness of positive predictions, but when the model never predicts positives, precision is undefined. Recall directly highlights the model's inability to find positives.

Full explanation →
43
MCQmedium

A company has a time series forecasting problem with daily sales data. The data shows both trend and seasonality. Which Amazon SageMaker built-in algorithm is most suitable?

A.K-Means
B.Linear Learner
C.DeepAR
D.XGBoost
AnswerC

DeepAR is a built-in algorithm for time series forecasting that handles trend and seasonality.

Why this answer

DeepAR is a supervised learning algorithm for time series forecasting that explicitly models both trend and seasonality using recurrent neural networks (RNNs). It is designed to handle multiple related time series, incorporate additional features like holidays or promotions, and produce probabilistic forecasts, making it the most suitable choice for daily sales data with trend and seasonal patterns.

Exam trap

The trap here is that candidates often choose XGBoost (Option D) because it is a powerful general-purpose algorithm, but they overlook that DeepAR is specifically designed for time series forecasting with trend and seasonality, whereas XGBoost requires manual feature engineering (e.g., lag variables, rolling statistics) to capture temporal patterns and does not natively produce probabilistic forecasts.

How to eliminate wrong answers

Option A is wrong because K-Means is an unsupervised clustering algorithm used for grouping data points based on similarity, not for forecasting time series with trend and seasonality. Option B is wrong because Linear Learner is a linear regression or classification algorithm that assumes independence of observations and cannot inherently capture temporal dependencies like seasonality or trend without extensive feature engineering. Option D is wrong because XGBoost is a gradient boosting algorithm primarily for tabular data and classification/regression tasks; while it can be used for time series with lag features, it is not purpose-built for sequential forecasting and lacks native support for probabilistic outputs and seasonal patterns.

Full explanation →
44
MCQhard

A data scientist runs a SageMaker training job that fails with the above error. The S3 bucket and object exist, and the IAM role has s3:GetObject permission. What is the MOST likely cause?

A.The S3 object was uploaded with incorrect checksum or is corrupted
B.The training instance does not have internet access
C.The S3 bucket has versioning enabled and the object version is not specified
D.The IAM role lacks kms:Decrypt permission for an encrypted S3 object
AnswerA

A corrupted file can cause size mismatch and zero-byte download.

Why this answer

The error indicates that SageMaker cannot read the S3 object, even though the bucket and object exist and the IAM role has s3:GetObject permission. The most likely cause is that the object was uploaded with an incorrect checksum or is corrupted, which prevents SageMaker from verifying the integrity of the data during the training job initialization. SageMaker uses ETag (MD5 checksum) validation when reading objects, and a mismatch triggers a failure.

Exam trap

The trap here is that candidates often assume permission or network issues are the root cause, but the error specifically points to a data integrity problem, which is a subtle but critical detail in SageMaker's S3 interaction.

How to eliminate wrong answers

Option B is wrong because SageMaker training jobs access S3 via AWS internal network endpoints, not the public internet; internet access is not required for S3 data retrieval. Option C is wrong because versioning does not require specifying an object version unless the object is explicitly requested by version ID; SageMaker defaults to the latest version if none is specified. Option D is wrong because the error message does not mention encryption or KMS; if the object were encrypted with SSE-KMS, the error would explicitly indicate a decryption permission failure, not a generic read failure.

Full explanation →
45
Multi-Selecthard

A company is building a real-time anomaly detection system for network traffic logs. The logs are ingested via Amazon Kinesis Data Streams and processed with an Amazon SageMaker endpoint for inference. The team needs to ensure that the inference results are stored durably and can be replayed for model retraining. The system must handle at least 10,000 records per second with low latency. Which three AWS services should the team use to build this architecture? (Select THREE.)

Select 3 answers
A.AWS Glue ETL
B.Amazon Kinesis Data Streams
C.Amazon Kinesis Data Analytics for Apache Flink
D.Amazon Kinesis Data Firehose
E.Amazon SageMaker
AnswersB, C, E

Kinesis Data Streams provides the ingestion layer with low latency and high throughput.

Why this answer

Amazon Kinesis Data Streams is the correct ingestion layer because it provides durable, real-time data streaming with the ability to handle over 10,000 records per second. It acts as the source of truth for network traffic logs, enabling low-latency processing and replay for model retraining.

Exam trap

The trap here is that candidates often confuse Kinesis Data Firehose with Kinesis Data Streams, assuming Firehose's simplicity and S3 integration make it suitable for real-time inference, but Firehose lacks the record-level replay and low-latency processing required for this use case.

Full explanation →
46
MCQhard

A company uses Amazon Redshift as a data warehouse. They need to load 50 TB of clickstream data from S3 into Redshift daily. The data arrives in 5-minute intervals as gzipped CSV files. The target table has a sort key and a distribution key. The load must complete within 2 hours. Which approach is MOST efficient?

A.Use AWS Glue to transform the data and write to Redshift using JDBC.
B.Use a staging table and then merge using a stored procedure.
C.Use a series of INSERT statements from a Lambda function.
D.Use the COPY command with a manifest file and gzip compression.
AnswerD

COPY is optimized for bulk loading from S3.

Why this answer

The COPY command is the most efficient way to load large volumes of data into Amazon Redshift because it uses the cluster's massively parallel processing (MPP) architecture to read data directly from S3 in parallel across all nodes. With a manifest file, you can specify multiple gzipped CSV files, and the gzip compression reduces network I/O and storage overhead. This approach can easily load 50 TB within 2 hours, especially when the target table has a sort key and distribution key, as COPY automatically leverages these for optimal data distribution and sorting during the load.

Exam trap

The trap here is that candidates may overcomplicate the solution by choosing AWS Glue or staging tables, not realizing that Redshift's COPY command is purpose-built for high-speed parallel ingestion from S3 and is the most efficient method for bulk data loads.

How to eliminate wrong answers

Option A is wrong because AWS Glue writing to Redshift via JDBC is a row-by-row or small-batch operation that cannot match the parallel throughput of the COPY command, and it would introduce unnecessary transformation overhead for already-structured CSV data. Option B is wrong because using a staging table and a stored procedure merge adds extra steps and complexity without improving load speed; the COPY command can directly load into the target table with proper sort and distribution keys, making a staging table redundant for this bulk load scenario. Option C is wrong because a series of INSERT statements from a Lambda function would be extremely slow and inefficient for 50 TB of data, as each INSERT is a single-row operation that cannot leverage Redshift's parallel processing, and Lambda has a 15-minute execution timeout that would require complex orchestration to handle the full load.

Full explanation →
47
MCQhard

A data scientist is building a recommendation system for an e-commerce platform using Amazon SageMaker. The system needs to provide personalized product recommendations based on user purchase history and product metadata. The dataset contains 10 million users and 1 million products. Which algorithm should the data scientist use as the core of the recommendation engine?

A.Linear Learner
B.XGBoost
C.K-Means
D.Factorization Machines
AnswerD

Factorization Machines handle sparse data well and are designed for recommendation tasks.

Why this answer

Factorization Machines (FM) are specifically designed for high-dimensional sparse data like user-item interactions, making them ideal for recommendation systems with 10 million users and 1 million products. FM can capture pairwise feature interactions (e.g., user-product affinities) efficiently using factorized parameters, which scales well to large datasets and supports personalized recommendations from purchase history and metadata.

Exam trap

The trap here is that candidates often pick XGBoost (B) because it is a powerful general-purpose algorithm, but they overlook that it cannot efficiently handle the extreme sparsity and pairwise interaction learning required for large-scale recommendation systems, which Factorization Machines are purpose-built for.

How to eliminate wrong answers

Option A is wrong because Linear Learner is a supervised learning algorithm for regression or classification that assumes linear relationships and cannot model complex feature interactions (e.g., user-product pairs) inherent in recommendation systems. Option B is wrong because XGBoost is a tree-based ensemble method that struggles with high-dimensional sparse categorical data (e.g., user and product IDs) and does not naturally handle pairwise interaction learning without extensive feature engineering. Option C is wrong because K-Means is an unsupervised clustering algorithm that groups similar users or products but cannot produce personalized recommendations based on user-item interactions or predict ratings for unseen pairs.

Full explanation →
48
MCQhard

A company uses Amazon Kinesis Data Streams to ingest real-time clickstream data from a website. The data is consumed by a Lambda function that writes records to an S3 bucket. Recently, the number of shards was increased from 2 to 4 to handle higher throughput. After the change, the Lambda function started processing records with increased latency and some records were being written out of order. What is the MOST likely cause?

A.The S3 bucket is not configured with versioning, causing overwrites.
B.The Lambda function is reading from the oldest sequence number, causing high IteratorAgeSeconds.
C.The Lambda function’s reserved concurrency is too low for the increased shard count.
D.The partition key used by the producer does not ensure that related records go to the same shard after resharding.
AnswerD

After resharding, the mapping of partition keys to shards changes. If ordering matters, the partition key must be chosen to keep related records together.

Why this answer

After resharding from 2 to 4 shards, the mapping of partition keys to shards changes. If the producer does not use a partition key that ensures related records (e.g., same user session) are routed to the same shard, records that were previously ordered within a shard may now be split across multiple shards. Since the Lambda consumer processes shards independently, records from the same logical sequence can arrive out of order, and the increased shard count can also cause higher latency if the consumer is not properly parallelized.

Exam trap

The trap here is that candidates often confuse increased shard count with a need for more concurrency (Option C), but the real issue is that resharding changes the partition-to-shard mapping, which can break ordering guarantees unless the producer explicitly handles the new hash range.

How to eliminate wrong answers

Option A is wrong because S3 versioning controls object overwrites and deletions, not the ordering or latency of records written by Lambda; out-of-order writes are caused by upstream data distribution, not S3 configuration. Option B is wrong because the Lambda function reads from the latest sequence number by default when using the Kinesis trigger, not the oldest; high IteratorAgeSeconds would indicate a slow consumer, not a configuration to read from the oldest record. Option C is wrong because reserved concurrency limits the maximum number of concurrent Lambda executions, but the default unreserved concurrency is usually sufficient for 4 shards; low concurrency would cause throttling (e.g., 429 errors), not out-of-order processing.

Full explanation →
49
MCQeasy

A data scientist needs to process a large volume of streaming data from IoT devices and store the results in Amazon S3 for further analysis. Which AWS service is most suitable for ingesting and processing this data in near real-time?

A.Amazon Redshift
B.AWS Glue
C.Amazon Kinesis Data Analytics
D.Amazon EMR
AnswerC

Kinesis Data Analytics processes streaming data in real-time.

Why this answer

Amazon Kinesis Data Analytics is the most suitable service because it can process streaming data from IoT devices in near real-time using SQL or Apache Flink, and directly output the results to Amazon S3. It is designed for continuous, low-latency ingestion and analysis of data streams, making it ideal for this use case.

Exam trap

The trap here is that candidates often confuse AWS Glue (batch ETL) with real-time processing, or assume Amazon Redshift can handle streaming ingestion via its COPY command, but neither supports true near real-time stream processing with sub-second latency.

How to eliminate wrong answers

Option A is wrong because Amazon Redshift is a data warehouse for batch analytics on structured data, not designed for real-time streaming ingestion or processing. Option B is wrong because AWS Glue is a serverless ETL service primarily for batch data preparation and cataloging, not for real-time stream processing. Option D is wrong because Amazon EMR is a big data platform for batch and stream processing using frameworks like Spark or Flink, but it requires more operational overhead and is not as directly optimized for simple near real-time ingestion and S3 output as Kinesis Data Analytics.

Full explanation →
50
MCQeasy

A company is building a binary classification model to predict customer churn. The dataset has 10,000 samples with 500 churners (positive class). The data scientist trains a logistic regression model and obtains an accuracy of 95%. However, the model predicts all customers as non-churn. Which metric should the data scientist use to evaluate the model's performance?

A.AUC-ROC
B.F1-score
C.Accuracy
D.Confusion matrix
AnswerB

F1-score balances precision and recall; with all negatives predicted, recall is 0, so F1 is 0, clearly showing poor performance on churners.

Why this answer

The F1-score is the harmonic mean of precision and recall, making it robust to class imbalance. Since the model predicts all customers as non-churn (accuracy 95% due to 9500 non-churners), precision for the positive class is undefined (0 true positives) and recall is 0, so the F1-score correctly reveals the model's failure to identify any churners.

Exam trap

AWS often tests the trap that candidates choose accuracy because it is high (95%), failing to recognize that accuracy is meaningless in imbalanced datasets when the model predicts only the majority class.

How to eliminate wrong answers

Option A is wrong because AUC-ROC can be misleading with severe class imbalance; a model that always predicts the majority class can still achieve a high AUC-ROC if the scores are well-separated, but it fails to capture the complete lack of positive predictions. Option C is wrong because accuracy is dominated by the majority class (95% non-churn) and gives a false sense of performance when the model never predicts the positive class. Option D is wrong because a confusion matrix is a visualization tool, not a single metric; while it would show zero true positives, the question asks for a metric to evaluate performance, and the confusion matrix itself is not a scalar metric.

Full explanation →
51
Multi-Selectmedium

Which TWO actions can help reduce overfitting in a decision tree model? (Choose 2.)

Select 2 answers
A.Prune the tree after training
B.Increase the maximum depth of the tree
C.Set a minimum number of samples per leaf
D.Increase the number of features considered at each split
E.Use all training data without validation
AnswersA, C

Pruning removes branches that have little predictive power, reducing overfitting.

Why this answer

Pruning the tree after training removes branches that have little predictive power, reducing overfitting by simplifying the model. This technique directly addresses the variance component of the bias-variance tradeoff, making the model generalize better to unseen data.

Exam trap

The MLS-C01 exam often tests the misconception that increasing model complexity (e.g., deeper trees or more features) always improves accuracy, when in fact it increases overfitting; candidates may incorrectly select options that add complexity instead of regularization.

Full explanation →
52
MCQeasy

A data scientist is exploring a dataset and finds that the correlation between two features is 0.95. What should the data scientist do to address multicollinearity before training a linear regression model?

A.Remove one of the two features
B.Apply L2 regularization
C.Standardize the features
D.Apply Principal Component Analysis
AnswerA

Removing one feature eliminates the high correlation.

Why this answer

Removing one of the highly correlated features reduces multicollinearity. Regularization (B) like Ridge can help but does not remove multicollinearity. Scaling (C) does not affect correlation, so it does not address multicollinearity.

PCA (D) can reduce multicollinearity by creating uncorrelated components, but it changes interpretability and is not the simplest solution.

Full explanation →
53
MCQeasy

A company is training a large language model on Amazon SageMaker using a single GPU instance. The training is taking too long. Which change would most likely reduce training time?

A.Use a larger instance with multiple GPUs and enable distributed training
B.Increase the instance memory
C.Decrease the batch size
D.Store the training data in S3 Glacier for faster access
AnswerA

A larger instance with multiple GPUs and distributed training allows parallel processing of data, reducing training time.

Why this answer

Using multiple GPUs in a distributed training job can parallelize work and reduce time. Option B is wrong because increasing instance memory may not help if the bottleneck is GPU compute. Option C is wrong because decreasing batch size often increases training time due to more updates per epoch.

Option D is wrong because S3 Glacier is designed for archival storage with slower retrieval, not faster access.

Full explanation →
54
MCQmedium

A data scientist is analyzing server logs stored in Amazon CloudWatch Logs. The above snippet shows three log entries. They want to count the number of 500 errors per minute using CloudWatch Logs Insights. Which query should they use?

A.fields @timestamp, status | filter status = 500 | stats count() by bin(1m)
B.fields @timestamp, @message | filter @message like /ERROR/ | stats count() by bin(1m)
C.fields @timestamp, @message | filter @message like /500/ | sort @timestamp desc
D.fields @timestamp, @message | filter @message like /500/ | stats count() by bin(1m)
AnswerD

Correctly filters for 500 status code and counts per minute.

Why this answer

The correct query is D. It reads the @timestamp and @message fields, filters log entries containing '500' in the message, and then counts the number of such entries per minute using stats count() by bin(1m). This directly matches the requirement to count 500 errors per minute.

Option A attempts to filter on a 'status' field, but the logs shown do not have a separate status field; the status code is embedded in the message. Option B filters for 'ERROR', but the logs use numeric status codes, not the word 'ERROR'. Option C sorts the results instead of aggregating, so it does not provide the count per minute.

Full explanation →
55
MCQmedium

A data scientist is exploring a dataset with 100 features. After generating pair plots, the scientist notices that many features have skewed distributions. Which transformation should the scientist apply to make the distributions more Gaussian-like for modeling?

A.Log transformation
B.Yeo-Johnson transformation
C.Standard scaling (z-score normalization)
D.Box-Cox transformation
AnswerB

Works for any real values.

Why this answer

(Yeo-Johnson transformation) is correct because it can handle both positive and negative values, making it suitable for datasets with skewed distributions that may include negative numbers. Option A (Log transformation) is wrong because it only works for strictly positive values. Option C (Standard scaling) does not change the shape of the distribution; it only centers and scales the data, so it does not fix skewness.

Option D (Box-Cox transformation) also requires positive values, limiting its applicability.

Full explanation →
56
MCQeasy

During EDA, a data scientist notices that a numeric feature 'age' has outliers beyond 3 standard deviations. What is the most appropriate first step?

A.Use the feature as-is in the model
B.Apply a log transformation to the feature
C.Remove all rows with outlier values
D.Investigate the source of the outliers
AnswerD

Understanding outliers guides proper handling.

Why this answer

The most appropriate first step when encountering outliers is to investigate their source (Option D). Outliers could indicate data entry errors, measurement issues, or genuine rare events. Blindly removing them (Option C) or transforming them (Option B) without understanding the cause may distort analysis.

Using the feature as-is (Option A) can bias models if outliers are erroneous. Investigation should precede any action.

Full explanation →
57
MCQeasy

A company is building a model to classify customer reviews as positive or negative. The dataset has 10,000 positive and 100 negative reviews. Which metric is most appropriate for evaluating model performance?

A.F1 score.
B.Accuracy.
C.Mean squared error.
D.AUC-ROC.
AnswerA

F1 score considers both precision and recall, good for imbalance.

Why this answer

(F1 score) is most appropriate because it balances precision and recall, making it suitable for imbalanced datasets where accuracy can be misleading (e.g., a model predicting all positive would achieve 99% accuracy). Option B (Accuracy) is misleading in imbalanced scenarios. Option C (Mean squared error) is for regression, not classification.

Option D (AUC-ROC) can be used but may provide an overly optimistic assessment in highly imbalanced data.

Full explanation →
58
MCQeasy

Refer to the exhibit. A SageMaker endpoint logs this error. What is the most likely cause?

A.The model is corrupted
B.There is a network connectivity issue
C.The input data type is incorrect
D.The input data has fewer features than the model expects
AnswerD

The error explicitly states shape mismatch: expected 10 features, got 8.

Why this answer

The error log indicates a mismatch between the number of features in the input data and the number of features the model was trained on. SageMaker's inference endpoint validates the input shape against the model's expected feature dimensions; when the input has fewer features, the model cannot perform the matrix operations required for prediction, resulting in this error. Option D correctly identifies this feature count mismatch as the root cause.

Exam trap

The trap here is that candidates may confuse a feature count mismatch with a data type error (Option C), because both involve input validation, but the error message specifically points to a shape or dimension mismatch rather than a type conversion failure.

How to eliminate wrong answers

Option A is wrong because a corrupted model would typically produce a model loading failure or a runtime error during deserialization, not a feature dimension mismatch during inference. Option B is wrong because a network connectivity issue would manifest as a timeout or connection refused error, not a model-level input validation error. Option C is wrong because an incorrect data type (e.g., string instead of float) would raise a type conversion error or a dtype mismatch, not a feature count error.

Full explanation →
59
MCQeasy

A machine learning team is developing a model to predict housing prices. They have a dataset with numerical features like square footage and number of bedrooms, and categorical features like neighborhood. Which preprocessing step is essential before training a linear regression model?

A.Normalize all numerical features to have zero mean and unit variance
B.Remove highly correlated features
C.One-hot encode categorical features
D.Apply Principal Component Analysis (PCA) to reduce dimensionality
AnswerC

Linear regression requires numerical input; one-hot encoding is needed for categorical variables.

Why this answer

One-hot encoding converts categorical features into binary columns, which linear regression requires. Option A is wrong because scaling is important but not the only essential step; encoding is needed first. Option B is wrong because removing correlated features can help with multicollinearity but is not essential for linear regression.

Option D is wrong because PCA reduces dimensionality but is optional and not a preprocessing step required before training.

Full explanation →
60
Multi-Selecthard

A data engineering team uses AWS Glue to run ETL jobs. They notice that jobs are taking longer to complete as data volume grows. They want to optimize performance without increasing cost significantly. Which THREE strategies should they consider?

Select 3 answers
A.Remove partitioning from the output
B.Partition the input data in S3
C.Use Amazon EMR instead of Glue
D.Convert input data to columnar format (e.g., Parquet)
E.Increase the number of DPUs (workers)
AnswersB, D, E

Enables parallel processing.

Why this answer

Partitioning input data in S3 (B) allows AWS Glue to use partition pruning, reading only the relevant subsets of data instead of scanning the entire dataset. This reduces I/O and processing time, directly addressing the performance degradation caused by growing data volumes.

Exam trap

The trap here is that candidates often assume adding more DPUs (E) is the only way to speed up Glue jobs, overlooking that data optimization strategies (partitioning and columnar formats) can yield similar or better performance gains without increasing cost.

Full explanation →
61
MCQmedium

A company is using AWS Glue ETL jobs to process data stored in Amazon S3. The jobs currently run sequentially and take too long. The data engineer wants to reduce job duration without rewriting the code. Which action is most effective?

A.Change the underlying EC2 instance type to a compute-optimized instance
B.Increase the number of DPUs (Data Processing Units) for the job
C.Convert the data from CSV to Parquet format
D.Enable job bookmarks to skip already processed data
AnswerB

More DPUs allow parallel execution, reducing job duration.

Why this answer

Increasing the number of DPUs (Data Processing Units) for the AWS Glue ETL job directly allocates more distributed computing resources, enabling parallel execution of the job's stages. This reduces the overall runtime without requiring any code changes, as Glue automatically distributes the workload across the additional DPUs.

Exam trap

The trap here is that candidates often confuse improving data format efficiency (Parquet) or incremental processing (job bookmarks) with solving a sequential execution bottleneck, when the direct solution is to increase parallelism via DPUs.

How to eliminate wrong answers

Option A is wrong because changing the EC2 instance type to compute-optimized only improves per-node performance, but the job still runs sequentially on a single node; without parallelization, the overall duration is limited by the sequential processing bottleneck. Option C is wrong because converting data from CSV to Parquet improves I/O efficiency and reduces data scan volume, but it requires rewriting the ETL code to read Parquet format and does not directly address the sequential execution issue. Option D is wrong because enabling job bookmarks only skips already processed data in incremental runs, which does not reduce the duration of the initial full run or the current sequential processing bottleneck.

Full explanation →
62
MCQeasy

A data analyst is investigating a dataset where the target variable is binary (0/1). The analyst wants to check for multicollinearity among the numerical features. Which statistical measure should the analyst use?

A.Variance Inflation Factor (VIF).
B.Mutual information between features and target.
C.Chi-square test of independence.
D.Pearson correlation coefficient between each pair of features.
AnswerA

VIF measures how much a feature is explained by other features.

Why this answer

Variance Inflation Factor (VIF) is the correct measure to check for multicollinearity among numerical features. VIF quantifies how much a feature is correlated with other features by calculating the ratio of variance of a model with multiple features to variance of a model with one feature. Option B (Mutual information) measures dependence between features and target, not among features.

Option C (Chi-square test) is for categorical variables. Option D (Pearson correlation) only measures pairwise linear relationships, not multicollinearity involving multiple features.

Full explanation →
63
MCQmedium

A data scientist is performing EDA on a dataset with many features. They suspect some features are redundant due to high pairwise correlations. Which technique can help identify groups of correlated features?

A.Use t-SNE to visualize feature relationships
B.Apply PCA and examine the loadings
C.Compute mutual information between each feature and the target
D.Use chi-square test for each pair
E.Create a correlation matrix and visualize with a heatmap
AnswerE

A correlation matrix heatmap clearly shows correlated feature groups.

Why this answer

Create a correlation matrix and visualize with a heatmap. This technique directly shows pairwise correlations between features, making it easy to identify groups of highly correlated (redundant) features. Option A is incorrect: t-SNE is a dimensionality reduction technique for visualization of high-dimensional data, but it does not quantify pairwise correlations between features.

Option B is incorrect: PCA reduces dimensionality by creating principal components that are linear combinations of original features; while loadings indicate feature contributions, they do not directly show pairwise correlations between original features. Option C is incorrect: Mutual information measures dependency between features and target, not between features themselves. Option D is incorrect: The chi-square test is used for testing association between categorical variables, not for continuous features or pairwise correlation analysis.

Full explanation →
64
Drag & Dropmedium

Drag and drop the steps to use Amazon SageMaker Feature Store for feature engineering in the correct order.

Drag steps to the numbered slots on the right, or tap a step then tap a slot.

Steps
Order
1Step 1
2Step 2
3Step 3
4Step 4

Why this order

Feature Store involves defining group, ingesting, querying, training, and maintaining.

Full explanation →
65
MCQhard

A financial services company is building a fraud detection model using Amazon SageMaker. The dataset has 10 million transactions, with 0.1% fraudulent. They train an XGBoost model with default hyperparameters. The model achieves 99.9% accuracy on the test set, but only catches 10% of actual fraud cases. The company wants to maximize the number of fraud cases caught while keeping the false positive rate below 5%. The data scientist has already tried adjusting the class weights and threshold, but the recall is still low. What should the data scientist do next?

A.Collect more data, especially fraudulent transactions, to balance the dataset
B.Use a different algorithm such as a balanced random forest or SMOTE with XGBoost
C.Apply PCA to reduce the number of features and prevent overfitting
D.Use a larger instance type to train for more epochs
AnswerB

Balanced random forest or SMOTE are designed to handle imbalanced datasets.

Why this answer

The model's high accuracy but low recall indicates it's predicting the majority class (non-fraud) most of the time due to extreme class imbalance (0.1% fraud). Default XGBoost does not handle such imbalance well. Using a balanced random forest (which under-samples the majority class during training) or SMOTE (synthetic oversampling of the minority class) with XGBoost can improve recall by making the model more sensitive to fraud.

Option A (collecting more data) may not help if the imbalance ratio stays the same. Option C (PCA) reduces dimensionality but does not address class imbalance. Option D (larger instance) only affects training speed, not model performance.

Full explanation →
66
MCQmedium

A data engineering team needs to ingest streaming data from thousands of IoT devices into Amazon S3 for near-real-time analytics. The solution must handle data that arrives in bursts and must be able to reprocess failed records automatically. Which combination of AWS services should the team use?

A.AWS Glue with Amazon S3
B.Amazon SQS with AWS Lambda
C.Amazon Kinesis Data Streams with AWS Lambda
D.Amazon DynamoDB Streams with AWS Lambda
AnswerC

Kinesis Data Streams can ingest bursty streaming data and retain it for replay; Lambda can process and load to S3.

Why this answer

Amazon Kinesis Data Streams is designed for real-time ingestion of large volumes of streaming data, such as from thousands of IoT devices, and can handle bursty traffic by scaling shards. AWS Lambda can be used as a consumer to process records in near-real-time, and Kinesis Data Streams supports automatic retries and checkpointing, enabling reprocessing of failed records without data loss.

Exam trap

The trap here is that candidates often confuse SQS with Kinesis, but SQS lacks the ordered, replayable stream semantics and high-throughput shard scaling needed for bursty IoT data ingestion and reprocessing.

How to eliminate wrong answers

Option A is wrong because AWS Glue is a batch ETL service, not designed for real-time streaming ingestion or automatic reprocessing of failed records in a streaming context. Option B is wrong because Amazon SQS with AWS Lambda is a message queue pattern that does not natively support ordered, replayable data streams or the high-throughput, bursty ingestion required from thousands of IoT devices; SQS lacks the shard-based parallelism and retention for reprocessing. Option D is wrong because Amazon DynamoDB Streams is designed to capture changes to DynamoDB tables, not for direct ingestion of raw IoT streaming data, and it does not provide the same level of throughput scaling or data retention for bursty IoT workloads.

Full explanation →
67
MCQmedium

A company needs to perform complex transformations on large datasets stored in Amazon S3 using Apache Spark. They want to minimize operational overhead. Which AWS service should they use?

A.Amazon EMR
B.Amazon EC2 with manually configured Spark
C.Amazon Athena
D.AWS Glue
AnswerA

EMR provides managed Spark clusters for complex transformations.

Why this answer

Amazon EMR is the correct choice because it is a managed big data platform that natively runs Apache Spark, allowing you to perform complex transformations on large datasets stored in Amazon S3 without provisioning or managing underlying infrastructure. EMR automatically handles cluster lifecycle, scaling, and tuning, minimizing operational overhead while providing full Spark compatibility.

Exam trap

The trap here is that candidates confuse AWS Glue's Spark-based ETL engine with a full-fledged Spark cluster, overlooking that Glue is optimized for simpler, serverless ETL jobs and lacks the fine-grained control and performance tuning capabilities of Amazon EMR for complex transformations.

How to eliminate wrong answers

Option B is wrong because manually configuring Spark on Amazon EC2 requires you to manage cluster setup, software installation, scaling, and fault tolerance, which increases operational overhead rather than minimizing it. Option C is wrong because Amazon Athena is a serverless interactive query service based on Presto, not Apache Spark, and it is designed for SQL-based ad-hoc queries, not for running complex Spark transformations. Option D is wrong because AWS Glue is a serverless ETL service that uses Apache Spark under the hood but abstracts away cluster management; however, it is primarily designed for batch ETL jobs with limited customization and does not provide the same level of control and performance tuning for complex Spark transformations as Amazon EMR.

Full explanation →
68
MCQhard

A company uses AWS Glue crawlers to populate the AWS Glue Data Catalog from Amazon S3. The data is partitioned by year/month/day/hour. The crawler runs every hour and adds new partitions. However, the data engineer notices that the crawler is taking longer to run as the number of partitions grows, and sometimes it misses new partitions. What is the most cost-effective and reliable way to address this?

A.Enable the crawler's partition index feature.
B.Manually add new partitions using ALTER TABLE ADD PARTITION in Athena.
C.Use the Athena MSCK REPAIR TABLE command after the crawler runs.
D.Increase the crawler's schedule to run every 30 minutes.
AnswerA

Partition indexes allow the crawler to efficiently discover new partitions without scanning the entire dataset.

Why this answer

Enabling the crawler's partition index feature allows AWS Glue to quickly find new partitions without re-scanning the entire table, reducing runtime and improving reliability. Option B is incorrect because manually adding partitions is error-prone and does not scale. Option C is incorrect because the MSCK REPAIR TABLE command is a manual step that does not automate the process.

Option D is incorrect because increasing the crawler's frequency does not address the underlying issue of scanning overhead and may lead to resource contention.

Full explanation →
69
MCQmedium

A company runs a daily ETL job that reads data from Amazon RDS, transforms it using AWS Glue, and writes the results to Amazon S3. The job started failing yesterday with the error: 'Rate exceeded'. What is the most likely cause and solution?

A.The Glue job is using too many DPUs; reduce the number of DPUs
B.The RDS database is overwhelmed by the number of connections; reduce the Glue job's parallelism or increase RDS instance size
C.The S3 bucket has reached its request rate limit; request a limit increase
D.Enable job bookmarks in the Glue job to process only new data
AnswerB

Rate exceeded errors often come from RDS when connection or IO limits are reached.

Why this answer

The 'Rate exceeded' error in an AWS Glue job reading from Amazon RDS typically indicates that the database is being overwhelmed by too many concurrent connections or queries. AWS Glue jobs can spawn multiple executors, each opening connections to RDS, and if the database's max_connections or IOPS limit is exceeded, RDS throttles requests. Reducing the Glue job's parallelism (e.g., setting the number of executors or DPUs lower) or scaling up the RDS instance (e.g., increasing instance size or provisioned IOPS) resolves the issue.

Exam trap

The trap here is that candidates often assume 'Rate exceeded' always refers to AWS API throttling (e.g., S3 or Glue API limits) rather than recognizing it as a database connection limit error, especially when the data source is RDS.

How to eliminate wrong answers

Option A is wrong because reducing DPUs might reduce parallelism but does not directly address the RDS connection limit; the error is from RDS, not from Glue resource limits. Option C is wrong because S3 bucket request rate limits are extremely high (thousands of requests per second) and typically cause 'SlowDown' errors, not 'Rate exceeded'; the error originates from the database, not S3. Option D is wrong because enabling job bookmarks only helps with incremental processing of new data, not with rate limiting; the job is failing due to connection overload, not because it is reprocessing old data.

Full explanation →
70
MCQmedium

A company uses Amazon Kinesis Data Streams to collect IoT sensor data. The stream has 4 shards. A consumer application reads from the stream using the Kinesis Client Library (KCL). The application processes records and stores them in Amazon DynamoDB. Recently, the data volume has increased, and the consumer is falling behind. Which action should the team take to increase the processing throughput?

A.Deploy additional consumer instances using the same application name.
B.Increase the write capacity of the DynamoDB table.
C.Increase the data retention period of the stream to 7 days.
D.Increase the number of shards in the Kinesis stream.
AnswerD

More shards provide more read capacity units and allow more parallel consumers.

Why this answer

Increase the number of shards in the Kinesis stream. Kinesis Data Streams throughput is determined by the number of shards; each shard provides 1 MB/s or 1000 records/s for reading. Since the consumer is falling behind, increasing shards directly increases the read throughput.

Option A is incorrect because adding more consumer instances without increasing shards will not improve throughput; KCL ensures each shard is processed by one worker, so extra workers are idle. Option B is incorrect because increasing DynamoDB write capacity may reduce throttling but does not address the root cause of low read throughput from Kinesis. Option C is incorrect because increasing the data retention period does not affect the rate at which data can be read.

Full explanation →
71
MCQeasy

A data scientist is using Amazon SageMaker to train a model. The training job is using a large dataset stored in S3. Which data input mode provides the FASTEST data loading for training?

A.FastFile mode
B.Augmented manifest file
C.File mode
D.Pipe mode
AnswerD

Pipe mode streams data directly from S3.

Why this answer

Pipe mode is the fastest data loading mode for SageMaker training because it streams data directly from S3 to the training algorithm via a FIFO pipe, bypassing disk writes. This eliminates the I/O overhead of downloading files to the local storage, enabling near-zero latency data ingestion for large datasets.

Exam trap

The trap here is that candidates often confuse 'FastFile mode' (which is fast but still disk-bound) with 'Pipe mode' (which is truly streaming), or they mistakenly think 'Augmented manifest file' is a data loading mode rather than a metadata file format.

How to eliminate wrong answers

Option A is wrong because FastFile mode, while optimized for high-throughput access, still requires data to be written to the local instance's Amazon Elastic Block Store (EBS) volume before training, adding disk I/O latency compared to streaming. Option B is wrong because an augmented manifest file is a metadata format for labeling jobs and data sources, not a data loading mode; it does not affect the speed of data transfer during training. Option C is wrong because File mode downloads the entire dataset from S3 to the local EBS volume before training starts, incurring significant download time and storage overhead, making it slower than streaming approaches.

Full explanation →
72
MCQmedium

A machine learning team is using SageMaker to train a model. They want to ensure that the training data is encrypted at rest in the S3 bucket and that the data is also encrypted during transit. Which configuration should they use?

A.Use client-side encryption and transfer data via HTTP
B.Use SSE-S3 encryption on the S3 bucket and enforce HTTPS
C.Use SSE-KMS encryption on the S3 bucket and disable HTTP
D.Use SSE-C encryption on the S3 bucket and HTTPS
E.Use no encryption on S3 but use HTTPS
AnswerB

SSE-S3 encrypts at rest; HTTPS encrypts in transit.

Why this answer

SSE-S3 provides server-side encryption at rest for objects in S3, and enforcing HTTPS ensures data is encrypted in transit using TLS. This combination meets the requirements for both encryption at rest and in transit without additional client-side complexity.

Exam trap

The trap here is that candidates often overthink encryption options and choose SSE-KMS or SSE-C, not realizing that SSE-S3 with HTTPS enforcement is the simplest and fully compliant solution for the stated requirements.

How to eliminate wrong answers

Option A is wrong because client-side encryption with HTTP does not encrypt data in transit (HTTP is plaintext), violating the transit encryption requirement. Option C is wrong because disabling HTTP entirely would break SageMaker's ability to communicate with S3 via HTTPS; SSE-KMS is valid for at-rest encryption but the statement 'disable HTTP' is impractical and unnecessary. Option D is wrong because SSE-C requires you to manage your own encryption keys, which adds operational overhead and is not the simplest configuration; HTTPS is correct for transit but SSE-C is not the best choice here.

Option E is wrong because no encryption at rest leaves data unencrypted in S3, failing the at-rest encryption requirement even though HTTPS provides transit encryption.

Full explanation →
73
MCQhard

A company is building a machine learning model to detect anomalies in industrial sensor data. The data is time-series with seasonal patterns. The data scientist wants to use Amazon SageMaker to train a model. Which algorithm is most suitable for this task?

A.Random Cut Forest (RCF)
B.K-Means
C.DeepAR
D.XGBoost
AnswerA

RCF is a SageMaker built-in algorithm for anomaly detection, suitable for time-series data.

Why this answer

Random Cut Forest (RCF) is the most suitable algorithm because it is designed for unsupervised anomaly detection on streaming and time-series data. It works by constructing an ensemble of random trees that isolate anomalies based on how quickly a data point can be separated from the rest, making it effective for detecting outliers in sensor data with seasonal patterns without requiring labeled training data.

Exam trap

The trap here is that candidates often confuse unsupervised anomaly detection with supervised forecasting or classification, leading them to pick DeepAR or XGBoost, but RCF is the only algorithm among the options specifically built for unsupervised anomaly detection on streaming data without requiring labels.

How to eliminate wrong answers

Option B (K-Means) is wrong because it is a clustering algorithm that groups data into clusters based on distance, not specifically designed for anomaly detection; it requires specifying the number of clusters (k) and does not inherently handle time-series seasonality or detect anomalies as outliers. Option C (DeepAR) is wrong because it is a supervised forecasting algorithm for time-series that predicts future values based on historical patterns, not for detecting anomalies in existing data; it requires a target variable and is not suited for unsupervised anomaly detection. Option D (XGBoost) is wrong because it is a supervised gradient boosting algorithm used for regression and classification tasks, requiring labeled data and feature engineering; it does not natively handle unsupervised anomaly detection or time-series seasonality without extensive preprocessing and labeling.

Full explanation →
74
MCQhard

A data engineer is performing exploratory data analysis on a dataset stored in Amazon S3 using AWS Glue DataBrew. The dataset contains a column 'age' with missing values. DataBrew's profile shows that the column has 5% missing values, a mean of 45, and a standard deviation of 15. Which imputation strategy should the engineer recommend to minimize bias if the missing data is Missing at Random (MAR)?

A.Replace missing values with the mean (45)
B.Remove rows with missing 'age' values
C.Replace missing values with the median
D.Use multiple imputation to generate several plausible values and combine results
AnswerD

Multiple imputation preserves the natural variability and provides valid statistical inferences under MAR.

Why this answer

Multiple imputation provides unbiased estimates under MAR by accounting for uncertainty and preserving relationships between variables. Option A is wrong because mean imputation reduces variance, distorts distributions, and can bias estimates. Option B is wrong because listwise deletion reduces sample size and can introduce bias if missingness is related to other variables.

Option C is wrong because median imputation is a single imputation method that does not account for the uncertainty due to missing data and may still introduce bias.

Full explanation →
75
MCQmedium

A data scientist wants to use AWS Step Functions to orchestrate a machine learning workflow including data preprocessing, training, and evaluation. Which SageMaker integration is best suited for this purpose?

A.Implement each step as an AWS Lambda function and call Step Functions
B.Use the SageMaker SDK with Step Functions service integrations
C.Use SageMaker Pipelines to define the workflow
D.Use AWS Batch to run the steps sequentially
AnswerB

Step Functions has built-in integrations for SageMaker training, processing, and endpoints.

Why this answer

AWS Step Functions has a native service integration with SageMaker, allowing you to directly call SageMaker API actions (e.g., CreateTrainingJob, CreateModel, CreateEndpointConfig, CreateEndpoint) without needing intermediate Lambda functions. This integration supports both synchronous and asynchronous invocations, making it ideal for orchestrating a multi-step ML workflow with built-in error handling, retries, and state management.

Exam trap

The trap here is that candidates often confuse SageMaker Pipelines (a dedicated ML pipeline service) with Step Functions, but the question explicitly asks for a SageMaker integration with Step Functions, not a replacement for it.

How to eliminate wrong answers

Option A is wrong because implementing each step as an AWS Lambda function adds unnecessary complexity, cold start latency, and a 15-minute execution timeout that may not accommodate long-running training jobs; Step Functions can directly invoke SageMaker APIs without Lambda. Option C is wrong because SageMaker Pipelines is a purpose-built orchestration service for ML workflows, but the question specifically asks about using Step Functions, and Pipelines is a separate service, not an integration with Step Functions. Option D is wrong because AWS Batch is designed for batch computing jobs and lacks the native state machine orchestration, error handling, and direct SageMaker API integrations that Step Functions provides.

Full explanation →

Page 1 of 23

Page 2

All pages

1234567891011121314151617181920212223

Practice MLS-C01 by domain

Target a specific domain to shore up weak areas.

Data EngineeringMachine Learning Implementation and OperationsModelingExploratory Data Analysis
See all domains with question counts →