MLS-C01 Exploratory Data Analysis Practice Test 11 — 15 Questions

Question 1

A data scientist is analyzing a dataset with a highly imbalanced binary target variable (1% positive class). The dataset contains 20 numerical features and 5 categorical features. The scientist wants to understand the relationship between each feature and the target, and also detect potential multicollinearity among numerical features. The scientist plans to use Amazon SageMaker Data Wrangler for EDA. Which combination of analyses should the scientist perform to BEST achieve these goals?

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

Accepted Answer

Use a histogram for each numerical feature colored by the target, a bar chart for each categorical feature showing target rate, and compute Variance Inflation Factor (VIF) for numerical features.. For an imbalanced binary target, histograms colored by the target and bar charts showing target rate per category directly reveal how each feature separates the classes. VIF is a standard measure for multicollinearity among numerical features, providing a clear threshold for concern. This combination efficiently addresses both the relationship with the target and collinearity, which are critical for subsequent modeling.

Answer

Use a box plot for each numerical feature grouped by the target, a pie chart for each categorical feature, and compute the correlation matrix for all features after one-hot encoding categorical variables.

Answer

Use a t-SNE plot to visualize the data in two dimensions, a heatmap of missing values, and compute the pairwise correlation matrix for numerical features.

Answer

Use a scatter plot matrix for numerical features and a bar chart for categorical features, and compute Pearson correlation for numerical features.

Question 2

A data engineer has a 40 GB Parquet dataset in Amazon S3 and wants to run interactive SQL for exploratory analysis, joining the fact table with two small dimension tables to inspect aggregate patterns before training a model. The queries must run without managing any cluster and should read only the needed columns. Which service should he use?

Accepted Answer

Amazon Athena, pointing a table at the S3 prefix and querying with standard SQL.. Amazon Athena is serverless and reads Parquet directly from S3 with column pruning, so it scans only the columns referenced by each query and requires no cluster to manage. It supports SQL joins across the fact and dimension tables. EMR and Redshift Spectrum both require clusters, and DynamoDB is not an analytical SQL engine over Parquet.

Answer

Amazon Redshift Spectrum querying the Parquet files through an external schema.

Answer

Amazon DynamoDB with a global secondary index on the join key.

Answer

Amazon EMR with a persistent Hadoop cluster running Hive.

Question 3

A data scientist is preparing a feature-engineering pipeline and must identify numeric features whose pairwise Pearson correlation exceeds 0.95 so one of each pair can be dropped. The dataset is 8 GB of Parquet in Amazon S3 with 120 numeric columns and 3 million rows, and the team wants to avoid writing custom distributed code. (Choose two.)

Accepted Answer

Use the SageMaker built-in PCA algorithm to reduce the 120 columns and drop the resulting principal components with low explained variance. The task needs pairwise correlation detection at scale without custom distributed code. Data Wrangler's correlation matrix analysis computes and visualizes pairwise correlations directly, while the SageMaker built-in PCA algorithm exposes redundancy among numeric features through explained variance and component loadings. Clarify, quick model, and Macie address model explanation, target importance, and data security respectively, none of which produce pairwise correlations.

Answer

Run a SageMaker Clarify bias and explainability job and read the feature attribution scores as a correlation measure

Answer

Use the SageMaker Data Wrangler quick model analysis to rank features by their target correlation

Answer

Use Amazon Macie to scan the S3 bucket and report columns whose values are statistically dependent

Question 4

A data scientist is analyzing a dataset with a highly imbalanced binary target variable (1% positive class). The dataset contains 20 numerical features and 5 categorical features. The scientist wants to understand the relationship between each feature and the target, and also detect potential multicollinearity among numerical features. The scientist plans to use Amazon SageMaker Data Wrangler for EDA. Which combination of analyses should the scientist perform to BEST achieve these goals?

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

Accepted Answer

Target encoding (mean encoding). Target encoding is the correct technique because it replaces each category with a statistic of the target variable, such as the mean, effectively grouping levels by their relationship to the target. This reduces dimensionality to a single feature and captures predictive information, which is beneficial for tree-based models. With proper regularization, it mitigates overfitting, making it ideal for high-cardinality categorical features.

Answer

Label encoding

Answer

Hashing trick (feature hashing)

Answer

One-hot encoding

Question 5

A data scientist is performing exploratory data analysis on a dataset stored in Amazon S3. The dataset contains missing values in several columns. The scientist wants to quickly visualize the pattern of missingness across all columns. Which approach is most efficient using AWS services?

Accepted Answer

Use AWS Glue DataBrew to create a missing value map visualization.. AWS Glue DataBrew is a visual data preparation tool that includes a missing value map as part of its data profiling. It can directly connect to S3 and generate the visualization without coding. This makes it the most efficient for quickly understanding missingness patterns. Other options require more setup or coding, which is less efficient for rapid EDA.

Answer

Write a Python script using pandas to compute the null counts and plot a heatmap with seaborn.

Answer

Use Amazon Athena to run a query that counts nulls per column, then visualize in Amazon QuickSight.

Answer

Use Amazon SageMaker Data Wrangler to generate a missing value report.

Question 6

A financial analytics team stores daily transaction records in Amazon S3 as Parquet files. They need to explore the data interactively, calculate summary statistics, and create scatter plots of transaction amount versus customer age, all without managing servers or copying data into a separate analytics database. Which approach BEST meets these requirements?

Accepted Answer

Use an Amazon SageMaker Studio notebook with the Python SDK to read the Parquet files directly from S3 using pandas, then compute summary statistics and generate plots with matplotlib or seaborn.. SageMaker Studio offers a managed Jupyter environment that can directly access S3 data using pandas and visualize it with standard Python plotting libraries, fulfilling the need for interactive EDA without server management. Other options either require provisioning clusters or involve non-interactive workflows that add unnecessary steps and operational overhead.

Answer

Use Amazon Athena to run SQL queries against the Parquet files, then visualize the results in Amazon QuickSight.

Answer

Launch an Amazon EMR cluster with Apache Spark and use a Zeppelin notebook to read the Parquet files from S3, compute statistics, and plot the data.

Answer

Load the Parquet files into Amazon Redshift using the COPY command, then use Redshift Spectrum to run queries and export results to CSV for plotting in Excel.

Question 7

A data scientist is using Amazon SageMaker Data Wrangler to explore a dataset containing missing values. They want to handle missing data appropriately before training a model. Which two actions can they take in Data Wrangler to address missing values? (Choose two.)

Accepted Answer

Use the 'Drop missing values' transform to remove rows with any missing values.. Data Wrangler offers transforms to handle missing values by either imputing them with statistics like mean, median, or mode, or by dropping rows with missing values. These are the two primary built-in methods for addressing missing data in the tool.

Answer

Use the 'One-hot encode' transform to create a binary indicator for missing values.

Answer

Use the 'Fill missing values' transform to replace missing values with the previous row's value.

Answer

Use the 'Impute missing values' transform to replace missing values with a constant value of zero.

Question 8

A data scientist is analyzing a dataset with a highly imbalanced binary target variable (1% positive class). The dataset contains 20 numerical features and 5 categorical features. The scientist wants to understand the relationship between each feature and the target, and also detect potential multicollinearity among numerical features. The scientist plans to use Amazon SageMaker Data Wrangler for EDA. Which combination of analyses should the scientist perform to BEST achieve these goals?

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

Accepted Answer

Amazon SageMaker Processing with a custom script using scikit-learn's CountVectorizer.. SageMaker Processing with a custom script using CountVectorizer is ideal for tokenizing text and computing word frequencies by class. It provides the flexibility to implement NLP preprocessing and can scale to handle the dataset size, while integrating with the SageMaker ecosystem.

Answer

Amazon Athena with a SQL query using regular expressions to split text into words and count them.

Answer

Amazon SageMaker Ground Truth for labeling the data.

Answer

Amazon Comprehend for sentiment analysis and key phrase extraction.

Question 9

A data scientist is performing exploratory data analysis on a dataset stored in Amazon S3. The dataset contains missing values in several columns. The scientist wants to use Amazon SageMaker Data Wrangler to handle missing values and also detect anomalies in a numerical feature. Which two actions should the scientist take in Data Wrangler to achieve these goals? (Choose two.)

Accepted Answer

Use the 'Handle Missing Values' transform to impute missing values with the median for numerical columns and with a constant for categorical columns.. To handle missing values and detect anomalies, the scientist should impute numerical missing values with the median (robust to outliers) and categorical with a constant, and use the Robust Z-Score anomaly detection method. The median imputation reduces bias from outliers, while Robust Z-Score uses median and MAD to reliably identify anomalies even in skewed distributions. Together they address both tasks.

Answer

Use the 'Handle Missing Values' transform to drop rows with missing values in the numerical feature of interest.

Answer

Use the 'Anomaly Detection' transform with the 'Standard Deviation' method on the numerical feature to identify outliers.

Answer

Use the 'Handle Missing Values' transform to impute missing values with the mean for numerical columns and with a constant for categorical columns.

Question 10

A data scientist is performing EDA on a dataset stored in Amazon S3 and wants to quickly generate a comprehensive profile report that includes summary statistics, distribution histograms, and correlation matrices with minimal coding. Which AWS service should the scientist use?

Accepted Answer

Amazon SageMaker Data Wrangler. Amazon SageMaker Data Wrangler is specifically designed for exploratory data analysis and provides built-in visualizations such as histograms, scatter plots, and correlation matrices with minimal coding. It allows data scientists to quickly profile data stored in Amazon S3 and integrate the analysis into SageMaker workflows.

Answer

AWS Glue DataBrew

Answer

Amazon QuickSight

Answer

Amazon Athena

Question 11

A machine learning engineer is preparing a tabular dataset for a fraud-detection model. The 'transaction_amount' feature is heavily right-skewed with a long tail, and the target 'is_fraud' has a base rate of 0.4%. The engineer applies Amazon SageMaker Data Wrangler's 'Robust Standard Deviation' scaler to transaction_amount and then trains a model. After deployment, the model performs well on the training distribution but poorly on recent transactions. Which explanation best accounts for this outcome?

Accepted Answer

Robust Standard Deviation scaling was a poor fit for a right-skewed feature, leaving extreme values under-normalized so the model over-relied on them and failed when the tail shifted.. Robust Standard Deviation scaling uses median and IQR, so it tolerates outliers but does not reshape a heavy right tail; extreme values remain relatively large after scaling. A model trained on that representation can become overly sensitive to tail values, and when the production distribution's tail moves, performance drops. Log or quantile transformation addresses the skew itself, which is the appropriate remedy before scaling.

Answer

Robust Standard Deviation scaling removed the skew by transforming values to ranks, destroying the magnitude information the model needs at inference.

Answer

The scaler's median and IQR were computed on the training data and stored as a transform, so the model expects that fixed scaling, but recent transactions fall outside the original distribution.

Answer

The 0.4% fraud base rate means accuracy is meaningless, so the poor performance is simply an artifact of class imbalance rather than any scaling decision.

Question 12

A data scientist is performing EDA on a dataset of customer transactions stored in Amazon S3. The dataset contains a column 'product_category' with over 500 unique values. She wants to reduce the cardinality by grouping rare categories into a single 'Other' category before training a model. Which approach should she take using Amazon SageMaker Data Wrangler?

Accepted Answer

Use the 'Handle Categorical' transform with the 'Group rare categories' option.. The 'Handle Categorical' transform in SageMaker Data Wrangler is designed to manage categorical features, including grouping rare categories. It allows specifying a threshold to combine infrequent values into a single category, thus reducing cardinality. This is essential before model training to avoid overfitting and high-dimensional feature spaces. The other transforms either increase dimensionality or do not address the grouping requirement.

Answer

Use the 'One-hot encode' transform.

Answer

Use the 'Encode Categorical' transform with ordinal encoding.

Answer

Use the 'Replace missing values' transform with a constant.

Question 13

A data scientist is exploring a sensor dataset with 30 numeric features. She computes the pairwise Pearson correlation matrix and finds that two features have a correlation coefficient of 0.97, yet a scatter plot of those two features shows a clear curved, non-linear relationship. She wants to quantify how strongly the two features are associated without assuming linearity. Which approach should she take?

Accepted Answer

Compute the Spearman rank correlation between the two features.. Spearman rank correlation measures monotonic association regardless of whether the relationship is linear, so it correctly captures the strong curved dependence between the two sensor features. Pearson correlation, trimming outliers, and standardization all remain tied to linear association, and variance inflation factor addresses multicollinearity rather than the strength of a bivariate non-linear relationship.

Answer

Standardize both features to zero mean and unit variance and recompute the Pearson correlation.

Answer

Recompute the Pearson correlation after removing rows where the two features exceed two standard deviations.

Answer

Compute the variance inflation factor for each feature and drop the one with the higher value.

Question 14

A data scientist is running exploratory data analysis on a 40 GB Parquet dataset stored in Amazon S3. She wants to compute per-column summary statistics and correlation matrices without downloading the full dataset to the SageMaker Studio instance's local storage, and she wants results in a single notebook cell. Which approach should she use?

Accepted Answer

Use the SageMaker Data Wrangler or PySpark processing job with Spark's summary() and corr() functions to compute statistics and correlations, writing the output to S3 for the notebook to read.. Distributed compute is the key requirement here: a 40 GB dataset cannot be held in a notebook's memory, so the analysis must run where the data lives. Spark-based processing in SageMaker reads Parquet directly from S3, computes summary statistics and correlations in parallel, and returns only the small result set, satisfying both the scale and the single-cell interaction constraints.

Answer

Load the Parquet files into a pandas DataFrame with pd.read_parquet, then call DataFrame.describe() and DataFrame.corr().

Answer

Query the S3 data with Amazon Athena and use Athena's built-in statistics, then pull the results into the notebook via the Athena API.

Answer

Use the AWS Glue DataBrew profile job and export the resulting profile JSON to S3, then load it into the notebook.

Question 15

A data scientist is performing EDA on a dataset stored in Amazon S3 and wants to quickly understand the distribution of a numerical feature. The scientist prefers a visual, interactive approach without writing code. Which AWS service should they use?

Accepted Answer

Amazon QuickSight. Amazon QuickSight is the correct choice because it is a fully managed BI service that enables interactive, no-code visualizations, including histograms, directly from S3 data. The other services are either code-based, focused on data preparation, or lack built-in visualization capabilities for exploratory analysis.

Answer

AWS Glue DataBrew

Answer

Amazon Athena

Answer

Amazon SageMaker Data Wrangler