MLS-C01 Practice Exam 64 — 20 Questions

Question 1

Which TWO of the following are appropriate use cases for using Amazon SageMaker BlazingText? (Choose 2)

Accepted Answer

Text classification using supervised learning.. Option A (text classification using supervised learning) is correct because BlazingText implements a supervised mode that trains a classifier on labeled text, supporting both text classification and multi-label classification with fast GPU-accelerated training. Option C (learning word embeddings from a large text corpus) is correct because BlazingText's unsupervised mode implements Word2Vec (skip-gram and CBOW with negative sampling) to produce dense word vectors from unlabeled text. Option B is not appropriate because time series forecasting is handled by Amazon Forecast or SageMaker's DeepAR, not BlazingText. Option D is not appropriate because image classification requires a vision model such as SageMaker's Image Classification algorithm or a CNN, whereas BlazingText operates on text tokens. Option E is not appropriate because sequence-to-sequence translation is performed by neural machine translation models (e.g., SageMaker Seq2Seq or Transformer), not by BlazingText's classification/embedding algorithms.

Answer

Time series forecasting.

Answer

Classifying images.

Answer

Sequence-to-sequence translation.

Question 2

A data engineer is performing EDA on a dataset stored in Amazon S3 where one categorical column has thousands of distinct values. Summary statistics show the column is heavily imbalanced, with a few categories covering most rows and a long tail of rare categories. Before feature engineering, the engineer wants a quick visual and statistical summary of the category frequencies without writing custom code. Which approach is MOST suitable?

Accepted Answer

Generate a Data Quality and Insights Report in SageMaker Data Wrangler to inspect the column's category distribution and imbalance.. The Data Quality and Insights Report profiles categorical columns by reporting distinct values and frequency distributions, which directly exposes the heavy imbalance and long tail described. It runs from SageMaker Data Wrangler with no custom code, matching the engineer's requirement. Crawlers, access logs, and CloudWatch Logs Insights describe schema or operational activity rather than the internal value distribution of a column.

Answer

Use Amazon CloudWatch Logs Insights to query the category frequencies from the S3 objects.

Answer

Create an AWS Glue crawler and rely on the generated Data Catalog table schema to reveal category frequencies.

Answer

Enable Amazon S3 server access logging and analyze the request patterns for the objects.

Question 3

A data scientist is performing exploratory data analysis on a dataset stored in Amazon S3 and wants to detect and understand missing values before training a model. Which two actions are appropriate for this goal? (Choose two.)

Accepted Answer

Generate a data quality report with Amazon SageMaker Data Wrangler that summarizes missing-value counts and percentages per column.. Detecting and understanding missing values requires first quantifying how much is missing per column and then examining the pattern of missingness across rows and features. A data quality report provides the counts and percentages, while a missingness pattern view reveals whether the gaps are random or systematic. Premature imputation or row deletion erases that information, and type conversion for empty strings corrupts the data.

Answer

Delete every row that contains at least one missing value from the dataset immediately.

Answer

Replace every missing value with zero across all columns before any inspection.

Answer

Convert all columns to the string data type so missing entries can be represented as empty strings.

Question 4

A data scientist is analyzing a dataset of customer churn with a binary target variable. The dataset contains 20 features, including both numerical and categorical variables. The scientist wants to identify which features have the strongest relationship with the target and also detect potential non-linear associations. The scientist decides to compute the mutual information between each feature and the target. After computing mutual information scores, the scientist finds that a categorical feature with 50 unique categories has the highest score. What is the most appropriate next step to ensure the mutual information score is not misleading?

Accepted Answer

Compute the mutual information using cross-validation or a holdout set to account for overfitting due to high cardinality.. Mutual information scores can be inflated for features with many unique values because the estimator can fit noise. To obtain a reliable estimate, the scientist should compute mutual information using cross-validation or a holdout set. This method evaluates the score on unseen data, mitigating overfitting and providing a more accurate assessment of the feature's true relationship with the target.

Answer

Discretize the categorical feature into fewer bins to reduce the number of unique values before computing mutual information.

Answer

Apply one-hot encoding to the categorical feature and then recompute mutual information.

Answer

Use the chi-square test instead, as it is not affected by high cardinality.

Question 5

A data scientist has a 40 GB Parquet dataset in Amazon S3 partitioned by date. She must run interactive EDA queries that join it with a 5 GB reference table of product categories, but she wants to avoid provisioning any cluster and wants results in seconds rather than minutes. Which approach BEST meets these requirements?

Accepted Answer

Create an Amazon Athena workgroup, register the Parquet prefix and the reference table in the AWS Glue Data Catalog, and query with Athena SQL.. Athena is the serverless query service that uses the AWS Glue Data Catalog to query S3 data in place, supports Parquet columnar pruning and partition elimination, and can join the partitioned fact data with the smaller reference table. It requires no cluster provisioning and returns interactive results, matching every stated constraint.

Answer

Load both datasets into Amazon DynamoDB tables and run PartiQL queries from the console for the joins.

Answer

Launch an Amazon EMR cluster with Apache Spark, copy both datasets to HDFS, and run Spark SQL joins interactively.

Answer

Use AWS Glue crawlers and then query the data through Amazon Redshift Spectrum on an existing ra3 cluster.

Question 6

A data scientist is exploring a dataset with a categorical feature 'city' that has 100 unique values. Before encoding, the scientist wants to reduce the number of categories by grouping rare cities into an 'Other' category. Which approach should the scientist use to determine which cities are considered rare?

Accepted Answer

Group cities that appear in less than a certain percentage of the total rows, such as 1%.. To reduce high cardinality, the scientist should group categories that are infrequent. A common rule is to combine categories that appear in less than a small percentage of the data, such as 1%, into an 'Other' category. This reduces the number of unique values while preserving the most frequent categories, which are likely more informative and stable.

Answer

Group cities that have the lowest mean target value.

Answer

Group cities that have the highest variance in the target variable.

Answer

Group cities that are not present in the validation set.

Question 7

A data scientist is preparing a dataset for a binary classification model. During EDA, they notice that one numerical feature has a highly skewed distribution with a long right tail. They decide to apply a transformation to make the distribution more symmetric. Which transformation is MOST appropriate for reducing right skewness?

Accepted Answer

Logarithmic transformation. A logarithmic transformation is the most effective for reducing strong right skewness because it compresses large values more than small ones, pulling the tail in and making the distribution more symmetric. Standardization and min-max normalization do not alter skewness, and square root is weaker than log for highly skewed data.

Answer

Min-max normalization

Answer

Standardization (z-score normalization)

Answer

Square root transformation

Question 8

A data scientist is working on a multi-class classification problem with 10 classes. The model outputs probabilities and the scientist wants to evaluate the model's ability to rank classes correctly. Which metric is most appropriate?

Accepted Answer

Area Under the ROC Curve (AUC-ROC). AUC-ROC is the most appropriate metric because it evaluates the model's ability to rank classes correctly by measuring the trade-off between true positive rate and false positive rate across all thresholds. For multi-class problems, AUC-ROC can be computed using one-vs-rest or one-vs-one strategies, and it directly assesses ranking quality independent of a specific decision threshold.

Answer

F1 score

Answer

Accuracy

Answer

Log loss

Question 9

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?

Accepted Answer

Visualize the missingness using a heatmap or bar chart.. Visualizing missingness with a heatmap (e.g., missingno.matrix) or bar chart (missingno.bar) reveals patterns such as whether missing values are random, clustered by row/column, or correlated across features. This diagnostic step informs the choice of imputation strategy — for example, MCAR vs. MAR vs. MNAR — before any values are filled or dropped. It is the standard first step in missing-data exploration.

Answer

Compute the correlation matrix of the features with missing values.

Answer

Drop all rows with any missing values.

Answer

Impute all missing values with the mean of each column.

Question 10

A company uses Amazon SageMaker Data Wrangler to perform exploratory data analysis. They want to detect outliers in a numerical column using the Interquartile Range (IQR) method. Which transformation should they apply in Data Wrangler?

Accepted Answer

Handle outliers. In Amazon SageMaker Data Wrangler, the 'Handle outliers' transformation is specifically designed to detect and treat outliers, and it supports the Interquartile Range (IQR) method as one of its detection strategies. Applying this transformation allows the user to identify outliers in a numerical column using IQR and choose how to handle them (e.g., remove or cap).

Answer

Impute

Answer

Normalize

Answer

Binning

Question 11

A data scientist is performing exploratory data analysis on a large dataset stored in Amazon S3. The dataset contains missing values in several columns. The scientist wants to understand the pattern of missingness to decide on an imputation strategy. Which two techniques are appropriate for identifying and visualizing missing data patterns? (Choose two.)

Accepted Answer

Create a missingno matrix plot to visualize the location of missing values across the dataset.. A missingno matrix plot visualizes the exact locations of missing values, making patterns like block missingness immediately apparent. Computing pairwise correlations of missingness indicators quantifies whether missing values in different columns tend to occur together, revealing systematic relationships. Together, they provide both visual and quantitative insights into missing data patterns, which are essential for choosing an imputation strategy.

Answer

Train a generative adversarial network (GAN) to impute missing values and then analyze the imputed values.

Answer

Perform a t-test comparing the distribution of a numerical feature for rows with and without missing values in another column.

Answer

Use a bar chart of the mean values of each column to infer missing data patterns.

Question 12

During EDA on a fraud-detection dataset, a data scientist finds that the 'transaction_amount' feature has a Pearson correlation of 0.02 with the binary fraud label, yet a gradient boosted tree model trained on it achieves high AUC. The scientist also notices the fraud rate rises sharply only for amounts above $9,000. Which conclusion best explains this?

Accepted Answer

The relationship is nonlinear and concentrated in the tail, so Pearson correlation understates the signal while a tree can isolate the high-amount region with a split.. A near-zero Pearson coefficient only rules out a linear association across the full range, while the observed fraud concentration above $9,000 is a nonlinear, threshold-shaped relationship. Gradient boosted trees search for optimal split points, so they can isolate the high-amount region and achieve high AUC even though the linear correlation is tiny.

Answer

The label must be imbalanced, and class imbalance always deflates Pearson correlation, so the coefficient cannot be trusted for any feature.

Answer

Tree-based models cannot use continuous features, so the AUC must come from other columns and the amount feature is irrelevant.

Answer

The feature is noise and should be dropped, because a near-zero Pearson correlation proves it carries no predictive signal for the label.

Question 13

A company has a dataset with a large number of missing values in several columns. The data scientist wants to impute missing values without introducing bias. Which approach should be used?

Accepted Answer

Use iterative imputation (MICE) to model missing values. Iterative imputation (MICE) models each feature with missing values as a function of the other features, iterating until convergence. This preserves relationships between variables and avoids the bias introduced by simple single-value substitutions. It is the recommended approach when the goal is to impute without distorting the underlying distribution.

Answer

Remove rows with any missing values

Answer

Replace missing values with the mode of each column

Answer

Replace missing values with the mean of each column

Question 14

A data scientist is examining a dataset for a binary classification problem. The target variable has a 1:1000 imbalance. Which technique should be used to assess model performance during exploratory data analysis?

Accepted Answer

Area under the Precision-Recall curve. With a 1:1000 class imbalance, the positive class is extremely rare. The Area Under the Precision-Recall curve (AUPRC) focuses on the performance of the positive class and is sensitive to changes in precision and recall, making it a robust metric for imbalanced datasets. Unlike ROC AUC, which can be overly optimistic when negatives dominate, AUPRC provides a realistic assessment of model performance on the minority class.

Answer

F1 score

Answer

Area under the ROC curve

Answer

Cohen's kappa

Question 15

A data scientist is analyzing a dataset with missing values in 30% of the rows for the 'age' column. The data scientist decides to impute the missing values with the median of the observed 'age' values. What is a potential drawback of this approach?

Accepted Answer

The imputed values may reduce the variance of the 'age' distribution.. Imputing missing values with the median of the observed data artificially concentrates imputed values around the center of the distribution. This reduces the overall variance of the 'age' column because the imputed values do not reflect the natural spread of the data, potentially distorting downstream analyses like regression or clustering that rely on variance structure.

Answer

The imputation will introduce bias if the missing values are not random.

Answer

Imputation using median is computationally expensive for large datasets.

Answer

The imputed values will increase the variance of the feature, leading to overfitting.

Question 16

A team is building a product recommendation system using matrix factorization in Amazon SageMaker. They notice that the model's training loss decreases steadily but validation loss starts increasing after 5 epochs. What is the most likely cause?

Accepted Answer

Overfitting. In matrix factorization for recommendation systems, a decreasing training loss with an increasing validation loss after several epochs is a classic sign of overfitting. The model is memorizing the training data (including noise) rather than learning generalizable patterns, which degrades its performance on unseen validation data.

Answer

Underfitting

Answer

Not enough training data

Answer

Learning rate too high

Question 17

In exploratory data analysis, a data scientist notices that the distribution of a feature 'income' is heavily right-skewed. Which transformation is most appropriate to reduce skewness?

Accepted Answer

Log transformation.. Log transformation is the most appropriate technique to reduce right skewness in a feature like 'income' because it compresses the long tail of high values while expanding the lower end, making the distribution more symmetric. This is particularly effective for income data, which often follows a log-normal distribution, and is a standard preprocessing step in machine learning to improve model performance.

Answer

Standardization (z-score).

Answer

Square transformation.

Answer

Min-max scaling.

Question 18

A data scientist is deploying a model using Amazon SageMaker for real-time inference. The model is memory-intensive and requires a GPU. Which instance type should be selected for the endpoint?

Accepted Answer

p3.2xlarge. The p3.2xlarge instance is correct because it provides a GPU (NVIDIA Tesla V100) with high memory bandwidth, which is essential for memory-intensive deep learning models requiring GPU acceleration for real-time inference. SageMaker endpoints for GPU-based models must use instance types from the P or G families, as CPU-only instances like i3, c5, or r5 lack the parallel processing capabilities needed for efficient GPU inference.

Answer

i3.2xlarge

Answer

c5.2xlarge

Answer

r5.2xlarge

Question 19

A team is training a SageMaker built-in Object Detection model on a dataset of street scenes and wants to detect small traffic signs, which occupy a tiny fraction of each image. After several training runs the model detects large vehicles reliably but misses most small signs. The team has already verified that the annotation files are correct and the images are high resolution. Which change to the training configuration is most likely to improve small-object detection?

Accepted Answer

Raise the input image resolution and reduce the downsampling so feature maps retain more spatial detail.. When a detector finds large objects but misses small ones, the usual cause is that repeated downsampling has collapsed small regions below the resolution the feature maps can represent. Increasing input resolution and retaining finer-grained feature maps restores the spatial detail the detection heads need, whereas batch size, learning rate, and metric choice do not change what the network can spatially resolve.

Answer

Lower the learning rate and train for more epochs on the same configuration.

Answer

Increase the mini_batch_size while keeping the input resolution unchanged.

Answer

Switch the evaluation metric from mean average precision to accuracy and continue training.

Question 20

A data scientist is performing EDA on a dataset with 100 features. They want to reduce dimensionality by removing highly correlated features. Which TWO approaches are appropriate? (Choose TWO.)

Accepted Answer

Compute a correlation matrix and remove one feature from each pair with correlation >0.95.. Option C is correct because computing a correlation matrix (e.g., pandas .corr() or numpy.corrcoef) and dropping one feature from each pair whose absolute correlation exceeds 0.95 directly targets the stated goal of removing highly correlated features, reducing redundancy while preserving one representative per correlated group. Option D is correct because PCA transforms the 100 features into orthogonal principal components, and selecting the components that cumulatively explain 95% of the variance achieves dimensionality reduction while eliminating the multicollinearity that correlated features introduce. Option A is not appropriate because random forest feature importance ranks features by predictive contribution, not by correlation, so it does not specifically remove highly correlated features. Option B is not appropriate because VarianceThreshold removes low-variance (near-constant) features, which is unrelated to correlation-based redundancy. Option E is not appropriate because L1 regularization is applied during model training to shrink coefficients, not as an EDA preprocessing step for explicitly detecting and removing correlated features.

Answer

Use feature importance from a random forest to select top features.

Answer

Remove features with low variance using VarianceThreshold.

Answer

Apply L1 regularization (Lasso) during model training to zero out coefficients of correlated features.