MLS-C01 Exploratory Data Analysis Practice Test 12 — 15 Questions

Question 1

A data scientist is exploring a dataset with missing values in several numerical columns. The scientist wants to understand the pattern of missingness and decide whether to impute or drop missing values. The scientist suspects that missingness is related to the target variable. Which visualization technique is most appropriate to reveal the relationship between missingness and the target?

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

Accepted Answer

Missingness matrix (missingno matrix). A missingness matrix is the correct visualization because it explicitly shows which values are missing across the dataset. By sorting or grouping by the target variable, you can visually inspect whether missingness is associated with the target. This insight is crucial for deciding whether to impute missing values or treat missingness as a feature, and it directly addresses the scientist's suspicion.

Answer

Boxplot of each numerical feature

Answer

Correlation heatmap

Answer

Histogram of each numerical feature

Question 2

A data analyst is exploring a dataset of housing prices stored in Amazon S3. The dataset contains missing values in several columns, and the analyst wants to visualize the pattern of missingness to decide on an imputation strategy. Which AWS service can generate a data profile that includes missing value statistics and visualizations?

Accepted Answer

AWS Glue DataBrew. AWS Glue DataBrew is designed for data profiling and preparation, offering built-in analysis of missing values, distributions, and correlations. It provides a visual interface that quickly highlights data quality issues, making it the most suitable service for this task.

Answer

Amazon QuickSight

Answer

Amazon SageMaker Data Wrangler

Answer

Amazon Athena

Question 3

A data analyst is exploring a dataset in Amazon SageMaker Studio and wants to quickly generate summary statistics and visualizations for a DataFrame loaded in a Jupyter notebook. The analyst needs to understand the distribution of a numerical feature and the count of a categorical feature. Which built-in SageMaker Studio feature can be used to automatically generate these visualizations without writing additional code?

Accepted Answer

SageMaker Studio's 'Data Wrangler' integration in the notebook. SageMaker Studio integrates Data Wrangler, which can be launched from a notebook and import a pandas DataFrame. Data Wrangler's visual interface allows analysts to generate histograms, bar charts, and summary statistics automatically. This built-in integration eliminates the need for additional coding, making it the most efficient way to perform quick EDA on a DataFrame within Studio.

Answer

SageMaker Studio's 'Experiments' tracking

Answer

SageMaker Studio Notebook's 'Data Profiling' feature

Answer

SageMaker Data Wrangler quick analysis

Question 4

A data scientist is exploring a dataset with two numerical features, 'age' and 'income'. The scientist suspects a non-linear relationship. Which technique is most appropriate to visualize and quantify the relationship while accounting for potential non-linearity?

Accepted Answer

Calculate the Spearman rank correlation coefficient.. Spearman rank correlation is designed to detect monotonic relationships, which can be non-linear. It ranks the data before computing correlation, making it robust to non-linearity and outliers. For age and income, where the relationship might plateau or change direction, Spearman provides a more accurate quantification than Pearson. Visualization can be done with a scatter plot, but the quantitative measure is key.

Answer

Compute the Pearson correlation coefficient.

Answer

Perform a linear regression of income on age and examine the R-squared value.

Answer

Create a hexbin plot of age versus income.

Question 5

A data scientist is performing EDA on a dataset with a binary target variable and 200 features. The dataset has a significant class imbalance (95% negative, 5% positive). The scientist wants to identify features that are most predictive of the positive class. Which two techniques are MOST appropriate for this task? (Choose two.)

Accepted Answer

Use a chi-square test of independence for each categorical feature against the target.. Point-biserial correlation and chi-square tests are both appropriate for assessing the relationship between individual features and a binary target, even with class imbalance. They are simple, interpretable, and directly measure association without being influenced by the relative class sizes.

Answer

Calculate the variance inflation factor (VIF) for each feature and remove those with high VIF.

Answer

Perform principal component analysis (PCA) and retain the top 10 principal components.

Answer

Train a logistic regression model with L1 regularization and select features with non-zero coefficients.

Question 6

A data scientist is performing exploratory data analysis on a dataset containing both numerical and categorical features. The target variable is binary. The scientist wants to identify which features are most strongly associated with the target before building a model. Which two techniques are MOST appropriate for this initial analysis? (Choose two.)

Accepted Answer

Use a chi-squared test of independence between each categorical feature and the target.. To identify associations between features and a binary target, point-biserial correlation for numerical features and chi-squared tests for categorical features are both statistically sound and appropriate for initial EDA. They are simple to compute and interpret, providing a quick screening of relevant features.

Answer

Compute the variance inflation factor (VIF) for each feature to detect multicollinearity.

Answer

Perform principal component analysis (PCA) on the entire dataset and select the components with the highest variance.

Answer

Train a linear regression model with the binary target as the dependent variable and examine the p-values of the coefficients.

Question 7

A machine learning engineer is performing exploratory data analysis on a dataset with a categorical feature 'product_category' that has 50 unique values. The engineer wants to encode this feature for a linear model but is concerned about the curse of dimensionality and wants to avoid creating 50 separate binary columns. Which encoding technique should the engineer use?

Accepted Answer

Target encoding (mean encoding). Target encoding is effective for high-cardinality categorical features because it replaces categories with a single numeric value derived from the target, thus reducing dimensionality and often improving model performance. It is particularly useful for linear models where one-hot encoding would create too many features. However, it must be applied with cross-validation to prevent data leakage.

Answer

Binary encoding

Answer

Label encoding

Answer

One-hot encoding

Question 8

A data scientist is performing exploratory data analysis on a dataset with a highly imbalanced binary target variable (1% positive class). The scientist wants to understand the relationship between several numerical features and the target, and also assess the need for resampling. Which two techniques are MOST appropriate for this analysis? (Choose two.)

Accepted Answer

Calculate the class imbalance ratio and visualize it with a bar chart.. Stratified distribution plots and class imbalance ratio visualization are both essential for understanding an imbalanced dataset. They reveal how features differ between classes and quantify the imbalance, guiding decisions on resampling and feature selection without distorting the data.

Answer

Use Amazon SageMaker Clarify to compute feature importance and detect bias.

Answer

Compute the correlation between each numerical feature and the target variable.

Answer

Apply SMOTE to the dataset to balance classes before further analysis.

Question 9

A data scientist is doing EDA on a dataset for a binary classifier and suspects the training data does not represent the population the model will see in production. The scientist wants to detect and quantify distribution problems before modeling. Which two techniques should the scientist apply? (Choose two.)

Accepted Answer

Compute population stability index and chi-square tests between the training feature distributions and a recent production sample.. Detecting representativeness problems before modeling means measuring distribution differences and subgroup proportions directly. SageMaker Clarify's pre-training bias metrics quantify class imbalance and label bias across facets, and statistical comparisons such as population stability index and chi-square tests quantify per-feature distribution shift between training data and a production sample. Model-based techniques like feature importance or resampling operate after or beside the distribution question and do not measure it.

Answer

Compare the training dataset against a held-out production sample using SageMaker Clarify's post-training metrics and data drift detection.

Answer

Apply SMOTE to oversample the minority class until the training set is balanced.

Answer

Train a baseline model and inspect its feature importance ranking to identify which columns are least predictive.

Question 10

A data scientist at a retail company is performing exploratory data analysis on a dataset of 2 million customer transactions stored in Amazon S3 as Parquet files. The dataset includes customer_id, transaction_date, product_category, quantity, and unit_price. The scientist needs to quickly compute summary statistics such as mean, median, standard deviation, and correlation matrix for the numerical features without loading the entire dataset into memory. The scientist also wants to visually inspect the distribution of unit_price and the correlation between quantity and unit_price. Which two AWS services or tools can be used together to accomplish this? (Choose two.)

Accepted Answer

Amazon Athena to run SQL queries for summary statistics and create visualizations in Amazon QuickSight.. SageMaker Data Wrangler and Amazon Athena with QuickSight both allow scalable EDA on large S3 datasets without loading all data into memory. Data Wrangler provides built-in visualizations and statistics, while Athena with QuickSight offers SQL-based aggregation and visualization. The other options either miss correlation analysis, require a trained model, or involve unnecessary infrastructure.

Answer

Amazon Redshift Spectrum to query the data and use Redshift's built-in machine learning for statistics.

Answer

Amazon SageMaker Clarify to compute feature importance and generate partial dependence plots.

Answer

AWS Glue DataBrew to profile the dataset and generate a correlation matrix.

Question 11

A data scientist is working with a dataset that has a large number of features (over 500) and suspects that many are highly correlated. The scientist wants to reduce dimensionality for exploratory analysis and visualization, while preserving the structure of the data. The scientist also wants to identify which original features contribute most to the principal components. Which technique should the scientist use in Amazon SageMaker?

Accepted Answer

Use SageMaker's built-in PCA algorithm with the 'regular' mode and analyze the component weights.. SageMaker's built-in PCA algorithm in regular mode computes principal components using full SVD, providing component weights that indicate how much each original feature contributes to each principal component. This allows the scientist to reduce dimensionality while interpreting feature importance. The regular mode is more accurate than randomized for smaller datasets, making it ideal for this exploratory analysis.

Answer

Use SageMaker's built-in linear learner algorithm with L1 regularization to select features.

Answer

Use SageMaker's built-in k-means clustering algorithm to group features and then select representatives.

Answer

Use SageMaker's built-in t-SNE algorithm to embed the data into two dimensions and inspect feature loadings.

Question 12

A data scientist is analyzing a dataset with a timestamp column and wants to explore whether there is a weekly seasonality pattern in a numerical metric. The dataset contains two years of daily data. Which visualization is BEST for this purpose?

Accepted Answer

A box plot of the metric grouped by day of the week.. A box plot grouped by day of the week is the best choice because it directly compares the distribution of the metric for each weekday, making any consistent weekly pattern immediately apparent. It summarizes the data effectively and is robust to outliers, which is important for noisy daily data.

Answer

A heatmap of the metric with weeks on the y-axis and days of the week on the x-axis.

Answer

A scatter plot of the metric against the timestamp.

Answer

A line plot of the metric over time with a 7-day rolling average overlay.

Question 13

A data scientist is exploring a dataset with missing values in several numerical columns. The scientist wants to understand the pattern of missingness and decide whether to impute or drop missing values. The scientist suspects that missingness is related to the target variable. Which visualization technique is most appropriate to reveal the relationship between missingness and the target?

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

Accepted Answer

Amazon SageMaker Data Wrangler. SageMaker Data Wrangler includes a built-in data quality and insights report that automatically summarizes target label distribution, missing values, and feature statistics. It connects directly to S3 and requires minimal coding, making it ideal for this EDA task. Other services like Athena or QuickSight require more manual effort to achieve the same comprehensive view.

Answer

Amazon QuickSight

Answer

AWS Glue DataBrew

Answer

Amazon Athena with SQL queries

Question 14

A machine learning engineer is exploring a dataset with a timestamp column and wants to detect whether there is a weekly seasonality pattern in a sensor reading. The engineer plots the sensor reading over time but the pattern is not obvious due to noise. Which visualization technique is MOST effective for revealing the weekly seasonality?

Accepted Answer

A box plot of sensor readings grouped by day of the week.. Box plots grouped by day of the week aggregate the sensor readings by the seasonal period, making systematic differences across days visible even when the raw time series is noisy. This is a common and effective EDA technique for identifying weekly seasonality.

Answer

A histogram of sensor readings with 100 bins.

Answer

A line plot of the raw sensor readings over time with a 7-day moving average overlay.

Answer

A scatter plot of sensor reading versus time with a linear regression line.

Question 15

A data scientist is performing EDA on a dataset with a highly skewed numerical feature and several categorical features. She wants to prepare the data for a linear model and needs to identify which transformations or encodings are appropriate. (Choose two.)

Accepted Answer

Apply a log or Box-Cox transformation to the skewed numerical feature to reduce skew before fitting the linear model.. For a linear model, skewed numeric predictors should be transformed to reduce skew and then standardized so scales are comparable and regularization behaves correctly. Both steps address the model's distributional and scaling assumptions. Blanket one-hot encoding, dropping the feature, and rank replacement either introduce dimensionality problems, discard information, or are better suited to non-parametric methods.

Answer

Drop the skewed numerical feature entirely and rely only on the categorical features for the model.

Answer

Replace the skewed numerical feature with its rank order to preserve monotonic relationships.

Answer

One-hot encode all categorical features regardless of cardinality to make them usable by the linear model.