MLS-C01 Exploratory Data Analysis Practice Test 9 — 15 Questions

Question 1

A data scientist is performing feature engineering on a dataset with high cardinality categorical features (e.g., ZIP codes with thousands of unique values). Which technique is most effective for reducing dimensionality while preserving predictive power?

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

Accepted Answer

Target encoding. Target encoding replaces each high-cardinality category with a statistic (typically the mean of the target) computed from the training data, collapsing thousands of unique ZIP codes into a single numeric feature that preserves predictive signal. This dramatically reduces dimensionality compared to one-hot encoding while retaining the relationship between the category and the target.

Answer

Hash encoding

Answer

One-hot encoding

Answer

Label encoding

Question 2

A data scientist is using Amazon SageMaker Data Wrangler to explore a dataset. They notice that a feature has a very high correlation (0.95) with the target variable. What should they do to avoid overfitting?

Accepted Answer

Remove the feature from the dataset. A feature with 0.95 correlation to the target is likely to cause the model to rely too heavily on it, leading to overfitting, especially if the feature is not causally related or is a proxy for the target. Removing the feature is a direct way to reduce the model's dependence on a single highly correlated variable and mitigate overfitting. While regularization and dimensionality reduction can help, the question specifically asks what to do to avoid overfitting given this observation, and removing the feature is the most direct action.

Answer

Use L2 regularization in the model

Answer

Apply PCA to reduce dimensionality

Answer

Standardize the feature using StandardScaler

Question 3

A data scientist is exploring a dataset containing customer transactions. They want to create a feature that captures the average purchase amount per customer over the last 30 days. Which approach is most efficient in Amazon SageMaker Processing?

Accepted Answer

Use PySpark with window functions in SageMaker Processing. PySpark window functions in SageMaker Processing are the most efficient approach because they allow partitioning by customer and computing rolling 30-day averages in a distributed, parallelized manner without shuffling the entire dataset into a single node. Window functions like avg() OVER (PARTITION BY customer_id ORDER BY transaction_date ROWS BETWEEN 30 PRECEDING AND CURRENT ROW) handle the time-based aggregation natively and scale horizontally. This avoids the memory and performance bottlenecks of single-node pandas or Python loops.

Answer

Use Amazon Athena SQL query with GROUP BY

Answer

Use a Python script with a for loop to calculate per customer

Answer

Use pandas groupby and rolling functions

Question 4

A data scientist is working with a dataset that has imbalanced classes (1% positive). They want to explore the data before modeling. Which visualization technique is most appropriate to understand the distribution of features with respect to the target class?

Accepted Answer

Parallel coordinates plot. Parallel coordinates plot can show feature patterns for minority vs majority class in high dimensions. Option A is wrong because box plots are univariate and do not show interactions between features. Option C is wrong because histograms are univariate and do not show interaction. Option D is wrong because scatter plot matrices become cluttered with many features.

Answer

Box plots grouped by class

Answer

Histograms overlaid by class

Answer

Scatter plot matrix

Question 5

A data scientist is analyzing a dataset with high multicollinearity. Which TWO techniques can help identify and address multicollinearity?

Accepted Answer

Use Principal Component Analysis (PCA). Option E (Compute Variance Inflation Factor (VIF)) is correct because VIF directly quantifies multicollinearity: VIF = 1/(1 - R²ᵢ) for each predictor regressed on the others, with values above ~5 or 10 signaling problematic collinearity. Option D (Use Principal Component Analysis (PCA)) is correct because PCA transforms correlated features into orthogonal principal components, eliminating multicollinearity by construction and addressing it for downstream modeling. Option A (Plot a correlation matrix) only detects pairwise linear relationships and misses multivariate collinearity, so it is not one of the two techniques that both identify and address multicollinearity. Option B (Apply Lasso regression) performs regularization and feature selection but does not explicitly diagnose multicollinearity. Option C (Use Recursive Feature Elimination (RFE)) is a feature-selection wrapper that does not specifically target or measure multicollinearity.

Answer

Plot a correlation matrix

Answer

Apply Lasso regression

Answer

Use Recursive Feature Elimination (RFE)

Question 6

A data scientist ran an AWS Glue ETL job that failed with the error shown. What is the most likely cause?

Accepted Answer

The DataFrame does not have a column named 'age'. The error message indicates that the column 'age' is not found in the DataFrame, which only contains columns [id, name, salary]. Option A is incorrect because the error is about a missing column, not a header mismatch. Option C is incorrect because schema evolution would add a column, not cause a missing column error. Option D is incorrect because there is no indication of a data type issue; the error is about column existence.

Answer

The CSV file has a header mismatch

Answer

The schema is evolving incorrectly

Answer

The data type of 'age' is incompatible

Question 7

A data scientist queried an Athena table and got only one row back, but the CSV file is 1 MB. What is the most likely reason?

Accepted Answer

The table is partitioned but the partition is not correctly defined. The most likely reason is that the table is partitioned but the partition is not correctly defined, causing Athena to scan only one partition (or none) and return a single row. In Athena, partitions are virtual directories in S3, and if the partition metadata in the Glue Data Catalog does not match the actual S3 prefix structure, queries will not scan the intended data. This is a common misconfiguration when data is loaded with partitions but the partition projection or MSCK REPAIR TABLE is not run.

Answer

The CSV file contains only one row

Answer

The table is not an external table

Answer

Athena does not support CSV format

Question 8

A data scientist is working on a predictive maintenance project for a manufacturing company. Sensor data is collected every second from 100 machines and stored in an Amazon S3 bucket as Parquet files, partitioned by machine_id and date. The dataset is massive (10 TB) and contains over 2000 features per machine. The data scientist needs to perform exploratory data analysis to identify which features are most predictive of machine failure. They have access to Amazon SageMaker Studio with a SageMaker Data Wrangler flow. The initial data exploration is taking too long due to the volume of data. The data scientist wants to speed up the analysis without losing accuracy in feature selection. Which course of action is most appropriate?

Accepted Answer

Use SageMaker Data Wrangler to create a stratified sample by machine_id and date, then analyze the sample. SageMaker Data Wrangler supports stratified sampling, which preserves the distribution of machine failure across machine_id and date, allowing for faster exploratory data analysis while maintaining representativeness for feature selection. Option A is incorrect because distributed processing with EMR on the full dataset may still be slow and is unnecessary when sampling can capture the signal. Option B is incorrect because using only one machine's data loses cross-machine variability and may bias feature selection. Option D is incorrect because random sampling does not guarantee preservation of time series order or failure distribution, potentially compromising analysis accuracy.

Answer

Switch to using Amazon EMR with Spark to perform distributed feature selection on the full dataset

Answer

Reduce the data to a single partition by concatenating all files and use only one machine's data

Answer

Use Amazon Athena to query a random sample of rows from the dataset

Question 9

A data scientist wants to understand the distribution of a continuous feature before training a model. Which visualization is most appropriate?

Accepted Answer

Histogram. A histogram is the correct choice because it bins a continuous feature into intervals and plots the frequency of observations in each bin, directly revealing the shape, central tendency, spread, and skewness of the distribution. This is exactly the purpose of univariate distribution analysis prior to model training. Scatter plots, box plots, and bar charts serve different analytical goals and do not show the full frequency distribution of a continuous variable.

Answer

Scatter plot

Answer

Box plot

Answer

Bar chart

Question 10

A data scientist runs a logistic regression and obtains a model with 95% accuracy on the training set. However, the model performs poorly on the test set. Which exploratory data analysis step should have been performed to identify this issue?

Accepted Answer

Checking for class imbalance in the target variable. High training accuracy combined with poor test performance is a classic symptom of overfitting, and one of the most common causes in classification tasks is class imbalance in the target variable. When one class dominates, a model can achieve high overall accuracy by predicting the majority class while failing on the minority class, which is exactly what shows up as a large train/test gap. Checking class distribution during EDA would have revealed this before modeling.

Answer

Generating a correlation matrix of features

Answer

Log transformation of skewed features

Answer

Creating a heatmap of missing values

Question 11

A company's dataset contains a feature 'zip_code' with 500 unique values. The data scientist wants to use this feature in a linear model. Which EDA step is most important before feature engineering?

Accepted Answer

Compute the frequency of each zip code. Because zip codes are categorical with high cardinality, analyzing the frequency distribution helps decide how to group or encode them (e.g., target encoding). Option A is wrong because checking missing values is not the most important step for handling high cardinality; frequency distribution is key. Option C is wrong because histograms are for continuous variables, not categorical. Option D is wrong because correlation requires numeric features, and zip codes are categorical.

Answer

Check the proportion of missing values

Answer

Plot a histogram of the feature

Answer

Calculate the correlation between zip code and the target

Question 12

A data scientist uses Amazon SageMaker Data Wrangler to perform EDA on a large dataset stored in S3. The data scientist notices that the target variable is highly imbalanced. Which SageMaker Data Wrangler transform can be used to address this during data preparation?

Accepted Answer

SMOTE (Synthetic Minority Over-sampling). SMOTE (Synthetic Minority Over-sampling) is available in SageMaker Data Wrangler to generate synthetic samples for the minority class, thereby addressing class imbalance. Option A is wrong because standard scaling normalizes numeric features, not handling imbalance. Option B is wrong because principal component analysis (PCA) reduces dimensionality, not imbalance. Option D is wrong because one-hot encoding converts categorical features, not for imbalance.

Answer

Standard scaling

Answer

Principal component analysis (PCA)

Answer

One-hot encoding

Question 13

A data scientist is analyzing a dataset with missing values. Which THREE methods are appropriate for handling missing data during EDA and preprocessing?

Accepted Answer

Remove rows with any missing values. Option A is correct because listwise deletion (dropping rows with any missing values) is a standard, valid preprocessing technique when missingness is low or MCAR, ensuring downstream computations use only complete cases. Option B is correct because mean imputation replaces missing numeric entries with the column mean, a common and simple univariate imputation method that preserves the mean of the feature. Option E is correct because median imputation substitutes missing values with the column median, which is a robust alternative to the mean when the distribution is skewed or contains outliers. Option C is not appropriate as a general method because replacing missing values with 0 introduces a fixed artificial value that can bias statistics and is only valid in specific domain contexts, not as a default EDA/preprocessing step. Option D is incorrect because most modeling algorithms cannot handle missing values directly and ignoring them leads to errors or biased results, so missing data must be addressed during preprocessing.

Answer

Replace missing values with 0

Answer

Ignore missing values and proceed with modeling

Question 14

Refer to the exhibit. A data scientist is unable to read a CSV file from the S3 bucket 'my-bucket' using SageMaker. The IAM policy attached to the SageMaker execution role is shown. What is the most likely cause of the failure?

Accepted Answer

The bucket uses server-side encryption with AWS KMS (SSE-KMS) and the policy lacks kms:Decrypt permission. The most likely cause is that the S3 bucket uses server-side encryption with AWS KMS (SSE-KMS), and the SageMaker execution role's IAM policy lacks the kms:Decrypt permission. Even if the policy allows s3:GetObject, accessing an SSE-KMS encrypted object requires decrypt permissions on the KMS key. Without it, the read operation fails with an access denied error.

Answer

The policy does not allow the s3:GetObject action

Answer

The policy does not grant read access to the bucket

Answer

The policy does not include s3:ListBucket action

Question 15

Refer to the exhibit. A data scientist runs the AWS CLI command shown to explore the contents of an S3 bucket. The command returns an empty array. However, the data scientist knows there are objects larger than 1000 bytes in the bucket. What is the most likely reason for the empty result?

Accepted Answer

The AWS CLI is not configured with the correct region for the bucket. The command syntax is correct: `--query 'Contents[?Size > `1000`]'` uses backticks for the numeric literal as per JMESPath. The empty result indicates the CLI could not find matching objects. Since the data scientist knows objects larger than 1000 bytes exist, the most likely cause is that the CLI is configured with a default region different from the bucket's region. Running the command without an explicit `--region` parameter causes it to query the bucket in the wrong region, returning no results. Option A is incorrect because the backticks are valid. Option B is incorrect because `list-objects-v2` supports `--query`. Option C is incorrect because `--query` is supported. Option D is correct because a region mismatch would cause the CLI to look for the bucket in the wrong location, resulting in an empty array even though the bucket contains objects.

Answer

The query syntax is incorrect; backticks should not be used

Answer

The command should use list-objects instead of list-objects-v2

Answer

The --query parameter is not supported by list-objects-v2