MLS-C01 Practice Exam 51 — 20 Questions

Question 1

A data scientist is exploring a dataset with 50 features and a binary target. The data scientist computes the correlation matrix and finds that two features, X1 and X2, have a correlation coefficient of 0.95. Which TWO actions should the data scientist consider? (Choose 2.)

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Remove one of the highly correlated features from the dataset.. Option B is correct because a correlation coefficient of 0.95 between X1 and X2 indicates strong multicollinearity, and removing one of the redundant features reduces redundancy and potential instability in models such as linear regression without losing much information. Option C is correct because PCA transforms the original correlated features into uncorrelated principal components, effectively addressing multicollinearity while retaining most of the variance across the 50-feature set. Option A is not appropriate because log transformations address skewness or nonlinearity, not linear correlation between two features. Option D is incorrect because creating an interaction term between two already highly correlated features would exacerbate multicollinearity rather than resolve it. Option E is irrelevant because the scenario describes correlated features, not missing values, so imputation does not address the correlation issue.

Answer

Apply a log transformation to X1 and X2.

Answer

Create an interaction term between X1 and X2.

Answer

Impute missing values for X1 and X2.

Question 2

A company is using Amazon SageMaker to run a hyperparameter tuning job. The tuning job uses Bayesian optimization. Which THREE statements about Bayesian optimization are correct? (Choose THREE.)

Accepted Answer

It works well for continuous hyperparameters. Options B, C, and D are correct. Bayesian optimization uses past trial results to select hyperparameter combinations (C), it works well for continuous hyperparameters (B), and it typically finds optimal values in fewer trials than random search (D). Option A is false because Bayesian optimization can handle many hyperparameters, not just 5. Option E is false because Bayesian optimization generally requires fewer trials than grid search, not more.

Answer

It can only handle a maximum of 5 hyperparameters

Answer

It requires more trials than grid search to find optimal values

Question 3

A data scientist is building a regression model to predict energy consumption. The dataset includes features like temperature, humidity, day of week, and holiday flags. The scientist uses a linear regression model and obtains an R-squared of 0.85 on training and 0.40 on test. The scientist suspects the model is not capturing non-linear relationships. Which approach should the scientist use to capture non-linearity?

Accepted Answer

Add polynomial features (e.g., squared terms and interactions). (add polynomial features) captures non-linear relationships by introducing squared terms and interactions, allowing the linear model to fit curvature. Option A (PCA) reduces dimensionality but does not add non-linearity. Option B (L1 regularization using Lasso) reduces overfitting by shrinking coefficients but does not introduce non-linear terms. Option C (removing features with low correlation) may lose information and does not help with non-linearity.

Answer

Apply PCA to the feature set

Answer

Increase L1 regularization using Lasso

Answer

Remove features with low correlation to the target

Question 4

A company is using Amazon Kinesis Data Firehose to load streaming data into Amazon S3. The data is in JSON format, and they want to convert it to Parquet before storage. What should they configure?

Accepted Answer

Enable data format conversion in Firehose and specify a Glue table. Amazon Kinesis Data Firehose supports built-in data format conversion from JSON to Parquet or ORC. By enabling this feature and specifying an AWS Glue table that defines the schema, Firehose automatically converts incoming JSON records to Parquet before delivering them to the S3 destination. This eliminates the need for additional compute resources or post-processing steps.

Answer

Use an AWS Lambda function to transform the data

Answer

Run an AWS Glue ETL job after data is in S3

Answer

Use Kinesis Data Analytics for Apache Flink to convert the format

Question 5

A retail company streams point-of-sale transactions into an Amazon Kinesis Data Stream with four shards. A consumer application using the Kinesis Client Library processes records and writes them to Amazon S3. During holiday peaks the stream's WriteProvisionedThroughputExceeded metric rises and consumers fall behind, yet the PutRecord API is called with the same partition key for every transaction. The team wants to reduce throttling and improve consumer parallelism with minimal application change. Which action should a data engineer take?

Accepted Answer

Increase the number of shards in the stream and modify the producer to use a high-cardinality partition key such as the transaction ID.. The stream is throttling because every record uses the same partition key, concentrating writes on one shard. Adding shards raises aggregate write capacity, and switching to a high-cardinality key distributes records so no single shard saturates. This also allows more KCL workers to lease shards in parallel, improving consumer throughput with limited code change.

Answer

Enable enhanced fan-out on the stream and keep the single partition key so each consumer gets dedicated throughput.

Answer

Raise the stream's retention period to 168 hours and increase the consumer's MaxRecords parameter per GetRecords call.

Answer

Switch the producer to the Kinesis Producer Library and rely on its aggregation to raise throughput per shard.

Question 6

A data scientist is training a multiclass classification model to categorize support tickets into 50 categories. The dataset has 100,000 labeled tickets. The scientist uses a random forest classifier with 100 trees. The model achieves 90% accuracy on the test set, but the F1-score for some rare categories is below 0.1. The scientist wants to improve performance on rare categories without significantly reducing overall accuracy. Which approach should the scientist try?

Accepted Answer

Use class_weight='balanced' or compute custom class weights. Class weights (class_weight='balanced' in scikit-learn) reweight the loss function so that misclassifying rare categories contributes more to the objective, effectively telling the random forest to prioritize minority-class recall. This directly targets the low F1 on rare categories while typically preserving overall accuracy because the majority classes still dominate the data. Computing custom weights (e.g., inversely proportional to class frequency) gives finer control when 'balanced' is too aggressive.

Answer

Increase the maximum depth of trees

Answer

Reduce the number of trees to 50 to prevent overfitting

Answer

Switch to a one-vs-rest logistic regression model

Question 7

A financial services company is training a gradient boosting model on Amazon SageMaker to predict loan defaults. The dataset is highly imbalanced: only 1.5% of loans default. After training with the built-in XGBoost algorithm and default hyperparameters, the model achieves 98% accuracy but fails to identify almost any defaulters. A machine learning specialist needs to improve the model's ability to detect defaults while maintaining reasonable precision. Which approach is MOST appropriate?

Accepted Answer

Set the scale_pos_weight hyperparameter in the XGBoost algorithm to the ratio of negative to positive examples.. The core issue is class imbalance, and SageMaker's built-in XGBoost provides the scale_pos_weight hyperparameter specifically for this purpose. Setting it to the negative-to-positive ratio reweights the positive class, making the model more sensitive to defaults. This is a direct, efficient fix that requires no additional data processing and is well-documented for imbalanced classification.

Answer

Use SageMaker Automatic Model Tuning to optimize the area under the ROC curve (AUC) metric.

Answer

Apply SMOTE to the training data using a SageMaker Processing job, then train the model on the balanced dataset.

Answer

Increase the number of boosting rounds to 1000 and reduce the learning rate to 0.01.

Question 8

A data scientist is evaluating a linear regression model. Which TWO metrics are appropriate for evaluating the model's performance?

Accepted Answer

R-squared. R-squared (A) is correct because it quantifies the proportion of variance in the continuous target variable explained by the linear regression model, making it a standard goodness-of-fit metric for regression. Root Mean Squared Error (B) is correct because it measures the average magnitude of prediction errors in the target's units by taking the square root of the mean squared differences between predicted and actual values, directly assessing regression accuracy. Precision (C), AUC-ROC (D), and F1 score (E) are classification metrics: precision and F1 rely on true/false positives and negatives from discrete class predictions, and AUC-ROC evaluates the trade-off between true positive rate and false positive rate across classification thresholds, so none apply to a continuous-output linear regression model.

Answer

Precision

Answer

Area Under the ROC Curve (AUC-ROC)

Answer

F1 score

Question 9

A data scientist wants to deploy a PyTorch model for real-time inference. Which SageMaker deployment option provides the lowest latency for single-digit millisecond responses?

Accepted Answer

SageMaker Real-Time Inference endpoint. SageMaker Real-Time Inference endpoints keep compute instances continuously running behind a persistent HTTPS endpoint, so requests are served immediately without cold starts or queueing. This architecture delivers consistent single-digit millisecond latency, which is why it is the standard choice for latency-sensitive, always-on inference. Asynchronous and Serverless options introduce queuing or cold-start overhead that breaks the single-digit millisecond requirement.

Answer

SageMaker Asynchronous Inference

Answer

SageMaker Serverless Inference

Answer

SageMaker Batch Transform

Question 10

A data scientist is building a binary classification model to predict customer churn. The dataset is highly imbalanced, with only 5% of customers churning. The scientist evaluates several models using accuracy, precision, recall, and F1 score. Which metric is most appropriate for comparing model performance in this scenario?

Accepted Answer

F1 score. The F1 score is the harmonic mean of precision and recall, making it the most appropriate metric for imbalanced binary classification because it balances both false positives and false negatives. In a dataset with only 5% churn, accuracy is misleading (a model predicting 'no churn' always achieves 95% accuracy), while precision or recall alone ignore the other type of error. F1 provides a single score that reflects performance on the minority class, which is typically the class of interest in churn prediction.

Answer

Accuracy

Answer

Precision

Answer

Recall

Question 11

A company wants to perform automated hyperparameter tuning for a model. Which Amazon SageMaker feature should be used?

Accepted Answer

Amazon SageMaker automatic model tuning. Amazon SageMaker automatic model tuning (also known as hyperparameter tuning) is the correct feature because it automates the process of searching for the optimal combination of hyperparameters for a machine learning model. It uses algorithms like Bayesian optimization, random search, or Hyperband to efficiently explore the hyperparameter space and find the best-performing configuration based on a specified objective metric.

Answer

Amazon SageMaker Clarify

Answer

Amazon SageMaker Ground Truth

Answer

Amazon SageMaker Debugger

Question 12

A company uses SageMaker to train a model each night. The training data is stored in an S3 bucket with SSE-S3 encryption. The training job fails with an access denied error. Which configuration is needed?

Accepted Answer

Create an IAM role with S3 read access and assign it to the SageMaker training job. SageMaker training jobs assume an IAM role that grants permissions to access input data in S3 and write output artifacts. If the role lacks s3:GetObject on the training bucket, the job fails with Access Denied. Creating an IAM role with S3 read access and assigning it to the training job is the required configuration.

Answer

Configure the training job to run in a VPC with S3 VPC Endpoint

Answer

Enable SSE-KMS on the S3 bucket

Answer

Add a bucket policy allowing s3:GetObject for all principals

Question 13

A data scientist is training a random forest model for a binary classification task. The dataset has 100,000 samples and 500 features. The model is overfitting. Which TWO actions are MOST likely to reduce overfitting?

Accepted Answer

Reduce the maximum depth of each tree. Option B is correct because reducing the maximum depth of each tree (max_depth) limits how finely the trees can partition the training data, which constrains model complexity and reduces variance that causes overfitting. Option E is correct because increasing the minimum number of samples required to split an internal node (min_samples_split) prevents the trees from creating splits based on very small, noisy subsets, thereby regularizing the forest and improving generalization. Option A is not correct because adding more trees generally stabilizes predictions and reduces variance only up to a point, but it does not address the underlying overfitting of individual deep trees. Option C is not correct because increasing the number of features considered at each split (max_features) tends to make trees more correlated and more complex, which can worsen overfitting. Option D is not correct because using all features for each tree removes the randomness that decorrelates trees in a random forest, typically increasing overfitting and reducing the benefit of the ensemble.

Answer

Increase the number of trees in the forest

Answer

Increase the number of features considered at each split

Answer

Use all features for each tree

Question 14

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?

Accepted Answer

Create a correlation matrix and visualize with a heatmap. A correlation matrix visualized as a heatmap directly shows pairwise correlation coefficients between all features, making it easy to spot groups of highly correlated (redundant) features. This is the standard EDA technique for identifying multicollinearity.

Answer

Use t-SNE to visualize feature relationships

Answer

Apply PCA and examine the loadings

Answer

Compute mutual information between each feature and the target

Answer

Use chi-square test for each pair

Question 15

A data scientist is training a neural network for image classification. The training loss is not decreasing significantly, and the validation loss is high. Which TWO actions should the scientist take to address potential vanishing gradients?

Accepted Answer

Use ReLU activation functions in hidden layers. ReLU activation functions help mitigate vanishing gradients because they output a constant gradient of 1 for positive inputs, preventing the gradient from shrinking as it propagates backward through many layers. This avoids the exponential decay of gradients that occurs with saturating activations like sigmoid or tanh, enabling effective training of deep networks.

Answer

Increase the learning rate

Answer

Switch activation functions from ReLU to sigmoid

Answer

Remove dropout layers

Question 16

A machine learning engineer is using SageMaker to train an XGBoost model on a dataset with a severe class imbalance (1:1000). The goal is to maximize recall on the minority class. Which hyperparameter tuning strategy is MOST appropriate?

Accepted Answer

Set scale_pos_weight to the ratio of negative to positive samples. scale_pos_weight is XGBoost's dedicated hyperparameter for class imbalance; setting it to the ratio of negative to positive samples (roughly 1000 in this case) reweights the positive class's gradient contribution so the model prioritizes minority-class recall. This directly addresses the 1:1000 imbalance at the loss-function level, which is the most targeted strategy for maximizing recall on the minority class.

Answer

Set max_delta_step to a high value

Answer

Increase subsample ratio to 1.0

Answer

Set objective to 'binary:logistic' and tune max_depth

Question 17

A machine learning team needs to deploy a model that makes real-time predictions with latency under 100 ms. The model is a deep neural network with 500 MB of parameters. Which AWS service should they use?

Accepted Answer

Amazon SageMaker real-time endpoint. Amazon SageMaker real-time endpoints are purpose-built for low-latency inference, supporting GPU instances, model caching, and autoscaling. A 500 MB deep neural network requires substantial memory and compute, and SageMaker endpoints can serve it with sub-100 ms latency using appropriate instance types (e.g., ml.g4dn, ml.p3).

Answer

AWS Glue

Answer

AWS Lambda with a container image

Answer

Amazon EMR

Question 18

A machine learning team is using SageMaker to build a model. They need to track hyperparameter tuning experiments, compare results, and visualize metrics. Which SageMaker feature should they use?

Accepted Answer

SageMaker Experiments. SageMaker Experiments is the feature designed to track, organize, and compare machine learning experiments, including hyperparameter tuning jobs. It automatically captures inputs, parameters, metrics, and artifacts for each trial, and provides a UI to visualize and compare results across runs. This directly matches the team's need to track tuning experiments and visualize metrics.

Answer

SageMaker Ground Truth

Answer

SageMaker Model Monitor

Answer

SageMaker Hyperparameter Tuning

Answer

SageMaker Debugger

Question 19

A data scientist is building a regression model to predict house prices. The dataset contains 10 features, including 'number_of_bedrooms' and 'square_footage'. The scientist observes that the model has high variance. Which TWO actions are most appropriate to reduce overfitting? (Choose TWO.)

Accepted Answer

Reduce model complexity by using a simpler model. High variance means the model is overfitting the training data, so reducing model complexity (Option A) is appropriate because a simpler hypothesis class (e.g., fewer parameters, shallower trees, or lower-degree polynomials) has less capacity to memorize noise and thus generalizes better. Adding L2 regularization (Option B) is also correct because it penalizes large weights by adding a lambda * sum(w_i^2) term to the loss, shrinking coefficients and constraining the model's effective capacity, which directly combats overfitting. The remaining options are not appropriate: increasing training epochs (C) typically worsens overfitting by letting the model fit noise further, decreasing training data (D) reduces the signal available and usually increases variance, and adding more polynomial features (E) increases model capacity and thus exacerbates overfitting rather than reducing it.

Answer

Increase the number of training epochs

Answer

Decrease the amount of training data

Answer

Add more polynomial features

Question 20

A data scientist is evaluating a regression model. Which TWO metrics are appropriate for evaluating regression performance?

Accepted Answer

Root Mean Squared Error (RMSE). Root Mean Squared Error (RMSE) is correct because it measures the average magnitude of prediction errors in the same units as the target variable by taking the square root of the mean of squared differences between predicted and actual values, making it a standard regression metric. R-squared is also correct because it quantifies the proportion of variance in the dependent variable explained by the model, with values typically between 0 and 1 (or negative for poor fits), directly assessing regression goodness-of-fit. The F1 score, AUC, and precision are not appropriate here because they are classification metrics: F1 and precision rely on true/false positives and negatives from discrete class predictions, and AUC evaluates the trade-off between true positive rate and false positive rate across classification thresholds, none of which apply to continuous regression outputs.

Answer

F1 score

Answer

Area Under the ROC Curve (AUC)

Answer

Precision