Data Analysis practice questions

A retail company wants to test whether a new website layout increases the conversion rate compared to the current layout. They randomly assign visitors to either the control or treatment group. Which statistical test is most appropriate to compare the conversion rates?

Trap 1: Two-sample t-test

Incorrect. A two-sample t-test compares means of continuous data, not proportions.

Trap 2: ANOVA

Incorrect. ANOVA compares means among three or more groups.

Trap 3: Logistic regression

Incorrect. Logistic regression models the probability of conversion but is overkill for a simple two-group comparison; a chi-square test is more direct.

A
Two-sample t-test
Why wrong: Incorrect. A two-sample t-test compares means of continuous data, not proportions.
B
Chi-square test
Correct. A chi-square test compares observed frequencies across categories, suitable for conversion rates (binary outcome).
C
ANOVA
Why wrong: Incorrect. ANOVA compares means among three or more groups.
D
Logistic regression
Why wrong: Incorrect. Logistic regression models the probability of conversion but is overkill for a simple two-group comparison; a chi-square test is more direct.

Question 2easymultiple choice

A data analyst needs to identify the most frequently occurring value in a dataset. Which measure of central tendency should they use?

Trap 1: Standard deviation

Standard deviation measures dispersion, not central tendency.

Trap 2: Median

Median is the middle value when sorted.

Trap 3: Mean

Mean is the average, not the most frequent value.

A
Mode
Mode is the most frequently occurring value.
B
Standard deviation
Why wrong: Standard deviation measures dispersion, not central tendency.
C
Median
Why wrong: Median is the middle value when sorted.
D
Mean
Why wrong: Mean is the average, not the most frequent value.

Question 3mediummultiple choice

A retail company wants to predict future sales based on historical data. Which modeling approach is most appropriate if the data shows a clear seasonal pattern?

Trap 1: Linear regression

Linear regression does not inherently handle seasonality.

Trap 2: K-means clustering

Clustering is unsupervised and not for prediction.

Trap 3: Logistic regression

Logistic regression is for classification, not forecasting.

A
Linear regression
Why wrong: Linear regression does not inherently handle seasonality.
B
Time series analysis
Time series analysis explicitly models seasonal patterns.
C
K-means clustering
Why wrong: Clustering is unsupervised and not for prediction.
D
Logistic regression
Why wrong: Logistic regression is for classification, not forecasting.

Question 4hardmultiple choice

A data analyst is building a model to predict customer churn. The dataset has 10,000 records with 500 churned customers. The model predicts churn with 95% accuracy, but only identifies 10% of actual churners. Which metric best highlights this issue?

Trap 1: Accuracy

Accuracy is high (95%) but misleading due to imbalance.

Trap 2: F1 score

F1 combines precision and recall, but recall alone directly shows the issue.

Trap 3: Precision

Precision may be high if false positives are low, but it does not capture missed churners.

A
Accuracy
Why wrong: Accuracy is high (95%) but misleading due to imbalance.
B
F1 score
Why wrong: F1 combines precision and recall, but recall alone directly shows the issue.
C
Recall
Recall is low (10%), showing the model fails to detect churners.
D
Precision
Why wrong: Precision may be high if false positives are low, but it does not capture missed churners.

Question 5easymultiple choice

A data analyst needs to combine two datasets that have the same columns but different rows. Which operation should they use?

Trap 1: Concatenate

Concatenate can be ambiguous; in pandas, it can combine along rows or columns, but append is more specific.

Trap 2: Merge

Merge is for combining on a common key, not simply adding rows.

Trap 3: Aggregate

Aggregate summarizes data, not combine rows.

A
Concatenate
Why wrong: Concatenate can be ambiguous; in pandas, it can combine along rows or columns, but append is more specific.
B
Append
Append adds rows from one dataset to another with same columns.
C
Merge
Why wrong: Merge is for combining on a common key, not simply adding rows.
D
Aggregate
Why wrong: Aggregate summarizes data, not combine rows.

Question 6mediummultiple choice

A data analyst is performing a hypothesis test with a significance level of 0.05. The p-value obtained is 0.03. What should the analyst conclude?

Trap 1: Fail to reject the null hypothesis

p < alpha, so we reject null.

Trap 2: Accept the null hypothesis

We never accept null; we either reject or fail to reject.

Trap 3: The result is practically significant

Statistical significance does not imply practical significance.

A
Reject the null hypothesis
p < alpha indicates statistically significant result.
B
Fail to reject the null hypothesis
Why wrong: p < alpha, so we reject null.
C
Accept the null hypothesis
Why wrong: We never accept null; we either reject or fail to reject.
D
The result is practically significant
Why wrong: Statistical significance does not imply practical significance.

Question 7hardmultiple choice

A data scientist trains a regression model and observes high variance with low bias. Which technique is most appropriate to reduce variance?

Trap 1: Increase polynomial features

This would likely increase variance further.

Trap 2: Use a smaller training set

Smaller training set typically increases variance.

Trap 3: Remove correlated features

This may reduce variance but can increase bias; not the most direct method.

A
Apply Ridge regularization
Ridge adds penalty to coefficients, reducing overfitting and variance.
B
Increase polynomial features
Why wrong: This would likely increase variance further.
C
Use a smaller training set
Why wrong: Smaller training set typically increases variance.
D
Remove correlated features
Why wrong: This may reduce variance but can increase bias; not the most direct method.

Question 8easymultiple choice

A data analyst is cleaning a dataset and finds missing values in a categorical variable representing customer region. Which imputation method is most appropriate?

Trap 1: Drop rows with missing values

Dropping rows may discard valuable data; imputation is preferred when appropriate.

Trap 2: Mean imputation

Mean is for numeric data.

Trap 3: Median imputation

Median is for numeric data.

A
Drop rows with missing values
Why wrong: Dropping rows may discard valuable data; imputation is preferred when appropriate.
B
Mode imputation
Mode is appropriate for categorical variables.
C
Mean imputation
Why wrong: Mean is for numeric data.
D
Median imputation
Why wrong: Median is for numeric data.

Question 9mediummultiple choice

A data analyst needs to visualize the distribution of a continuous variable across different categories. Which chart type is most suitable?

Trap 1: Bar chart

Bar chart typically shows aggregate values, not distribution.

Trap 2: Histogram

Histogram shows distribution of one continuous variable without categories.

Trap 3: Scatter plot

Scatter plot is for two continuous variables.

A
Bar chart
Why wrong: Bar chart typically shows aggregate values, not distribution.
B
Histogram
Why wrong: Histogram shows distribution of one continuous variable without categories.
C
Scatter plot
Why wrong: Scatter plot is for two continuous variables.
D
Box plot
Box plot displays distribution across groups.

Question 10hardmultiple choice

A company is analyzing customer feedback sentiment. The dataset is highly imbalanced with 95% positive and 5% negative comments. Which technique should the analyst use to address class imbalance before modeling?

Trap 1: Use accuracy as the evaluation metric

Accuracy is misleading for imbalanced data; F1 or AUC are better.

Trap 2: Undersample the majority class

Undersampling can help but may discard useful data; SMOTE is more effective.

Trap 3: Oversample the majority class

This would increase imbalance further.

A
Use accuracy as the evaluation metric
Why wrong: Accuracy is misleading for imbalanced data; F1 or AUC are better.
B
Undersample the majority class
Why wrong: Undersampling can help but may discard useful data; SMOTE is more effective.
C
Oversample the majority class
Why wrong: This would increase imbalance further.
D
Use SMOTE
SMOTE generates synthetic minority samples to balance classes.

Question 11mediummulti select

Which TWO of the following are common assumptions of linear regression?

Trap 1: No multicollinearity

Incorrect. No multicollinearity is an assumption for multiple regression, not for simple linear regression, and even in multiple regression it is not always considered a 'common' assumption in the same sense.

Trap 2: Normality of the dependent variable

Incorrect. Linear regression assumes normality of residuals, not the dependent variable.

Trap 3: Homoscedasticity

Incorrect. Although homoscedasticity is an important assumption for unbiased standard errors, it is not selected here as it is often not considered one of the most common assumptions for basic linear regression.

A
Independence of observations
Correct. Independence of observations is a key assumption of linear regression; violations can lead to biased standard errors.
B
No multicollinearity
Why wrong: Incorrect. No multicollinearity is an assumption for multiple regression, not for simple linear regression, and even in multiple regression it is not always considered a 'common' assumption in the same sense.
C
Linearity of the relationship
Correct. Linearity is a fundamental assumption; the model must correctly specify a linear relationship.
D
Normality of the dependent variable
Why wrong: Incorrect. Linear regression assumes normality of residuals, not the dependent variable.
E
Homoscedasticity
Why wrong: Incorrect. Although homoscedasticity is an important assumption for unbiased standard errors, it is not selected here as it is often not considered one of the most common assumptions for basic linear regression.

Question 12hardmulti select

Which THREE of the following are appropriate methods to handle outliers in a dataset?

Trap 1: Binning continuous variables

Binning is a discretization method, not specifically for outlier handling.

Trap 2: Imputing outliers with the mean

Mean is sensitive to outliers; median is better.

A
Transforming the data using log transformation
Transformation can reduce the impact of outliers.
B
Removing the outlier records
Removal is acceptable if outliers are errors or not representative.
C
Capping the outlier values at a certain percentile
Capping limits extreme values.
D
Binning continuous variables
Why wrong: Binning is a discretization method, not specifically for outlier handling.
E
Imputing outliers with the mean
Why wrong: Mean is sensitive to outliers; median is better.

Question 13easymulti select

Which TWO of the following are examples of supervised learning algorithms?

Trap 1: K-means clustering

Unsupervised learning.

Trap 2: Principal component analysis (PCA)

Unsupervised dimensionality reduction.

Trap 3: Apriori algorithm

Unsupervised association rule mining.

A
Linear regression
Supervised regression algorithm.
B
K-means clustering
Why wrong: Unsupervised learning.
C
Principal component analysis (PCA)
Why wrong: Unsupervised dimensionality reduction.
D
Decision trees
Supervised classification/regression.
E
Apriori algorithm
Why wrong: Unsupervised association rule mining.

Question 14hardmultiple choice

A healthcare analytics team is building a predictive model to identify patients at high risk of readmission within 30 days of discharge. The dataset includes 50,000 patient records with 200 features, including demographics, vital signs, lab results, and historical admissions. The target variable is binary (readmitted or not). The team uses a logistic regression model and achieves an AUC of 0.72 on the test set. However, the model's calibration is poor: for patients predicted to have a 70% risk, the actual readmission rate is only 40%. The team wants to improve calibration without significantly reducing discrimination (AUC). The data scientist suggests applying Platt scaling. However, the team lead is concerned that Platt scaling may reduce the model's ability to rank patients correctly. Which of the following is the best course of action?

Trap 1: Remove poorly calibrated predictions by discarding all patients…

Discarding data reduces sample size and may introduce bias.

Trap 2: Ignore calibration because AUC is the only metric that matters for…

Calibration is important for decision-making; AUC alone can be misleading.

Trap 3: Switch to a random forest model, which inherently produces…

Random forests often have better calibration but may not necessarily preserve AUC; also, this changes the model entirely.

A
Remove poorly calibrated predictions by discarding all patients with predicted risk between 0.3 and 0.7.
Why wrong: Discarding data reduces sample size and may introduce bias.
B
Ignore calibration because AUC is the only metric that matters for readmission risk models.
Why wrong: Calibration is important for decision-making; AUC alone can be misleading.
C
Apply Platt scaling on a held-out validation set to recalibrate the predicted probabilities without refitting the original model.
Platt scaling is designed to improve calibration while maintaining AUC.
D
Switch to a random forest model, which inherently produces better-calibrated probabilities.
Why wrong: Random forests often have better calibration but may not necessarily preserve AUC; also, this changes the model entirely.

Question 15mediummultiple choice

A data analyst at a marketing firm is tasked with segmenting customers based on their purchasing behavior. The dataset contains 10,000 customers with features such as annual spend, frequency of purchases, recency of last purchase, and average order value. The analyst decides to use k-means clustering. After standardizing the features, the analyst runs k-means with k=3, k=4, and k=5, and computes the silhouette score for each: k=3: 0.45, k=4: 0.52, k=5: 0.48. The analyst also plots the elbow curve and observes that the within-cluster sum of squares (WCSS) decreases sharply from k=2 to k=4, then levels off. Based on these results, what is the most appropriate number of clusters?

Trap 1: k=2

Not evaluated, but elbow suggests k=4 is better.

Trap 2: k=3

Silhouette score is lower than k=4.

Trap 3: k=5

Silhouette score is lower than k=4.

A
k=4
Highest silhouette score and elbow point.
B
k=2
Why wrong: Not evaluated, but elbow suggests k=4 is better.
C
k=3
Why wrong: Silhouette score is lower than k=4.
D
k=5
Why wrong: Silhouette score is lower than k=4.

Question 16easymulti select

A data analyst is building a linear regression model to predict sales based on advertising spend across TV, radio, and newspaper channels. Which TWO diagnostics should the analyst perform to validate the model assumptions?

Trap 1: Durbin-Watson test for autocorrelation

Autocorrelation is not a core assumption for linear regression.

Trap 2: Variance inflation factor (VIF) for multicollinearity

Multicollinearity is a concern but not a core assumption.

Trap 3: Cook's distance to identify influential points

Influential points are diagnostics but not assumptions.

A
Durbin-Watson test for autocorrelation
Why wrong: Autocorrelation is not a core assumption for linear regression.
B
Q-Q plot to assess normality of residuals
Q-Q plot checks normality assumption.
C
Variance inflation factor (VIF) for multicollinearity
Why wrong: Multicollinearity is a concern but not a core assumption.
D
Cook's distance to identify influential points
Why wrong: Influential points are diagnostics but not assumptions.
E
Residual plots to check for homoscedasticity
Residual plots help verify constant variance.

Question 17hardmultiple choice

A data analyst is preparing a logistic regression model to predict customer churn. After examining the exhibit, which data quality issue should the analyst address first?

Network Topology

Trap 1: Duplicate customer IDs

All IDs appear unique.

Trap 2: Missing values in total_charges

No missing values are present.

Trap 3: Outliers in monthly_charges

Values like 99.65 are not extreme outliers.

A
Duplicate customer IDs
Why wrong: All IDs appear unique.
B
Missing values in total_charges
Why wrong: No missing values are present.
C
Inconsistent data in total_charges
The total_charges for the first customer is equal to monthly_charges, suggesting a calculation error.
D
Outliers in monthly_charges
Why wrong: Values like 99.65 are not extreme outliers.

Question 18mediummultiple choice

A healthcare analytics team is building a classification model to predict patient readmission within 30 days. The dataset contains 10,000 records with 30 features, including demographics, vital signs, lab results, and medication history. The target variable is imbalanced: 85% no readmission, 15% readmission. The team used logistic regression with default settings and achieved an accuracy of 85%, but the model predicted 'no readmission' for all patients. The lead analyst suspects the model is not learning due to class imbalance. The team has time to implement one corrective action before the next model review. Which action should the team take?

Trap 1: Remove features with low variance to reduce noise

Low variance features are not the primary issue; the model is not learning from the minority class.

Trap 2: Use accuracy as the evaluation metric to monitor improvement

Accuracy is misleading for imbalanced data; precision-recall or AUC should be used.

Trap 3: Switch to a random forest model with default settings

Without handling imbalance, random forest may still predict majority class.

A
Remove features with low variance to reduce noise
Why wrong: Low variance features are not the primary issue; the model is not learning from the minority class.
B
Apply SMOTE to oversample the readmission class
SMOTE generates synthetic samples, balancing the classes and allowing the model to learn from the minority class.
C
Use accuracy as the evaluation metric to monitor improvement
Why wrong: Accuracy is misleading for imbalanced data; precision-recall or AUC should be used.
D
Switch to a random forest model with default settings
Why wrong: Without handling imbalance, random forest may still predict majority class.

Question 19mediumdrag order

Drag and drop the steps to normalize a database table from 1NF to 3NF in the correct order.

Drag steps to the numbered slots on the right, or tap a step then tap a slot.

Steps

Order

1Step 1

2Step 2

3Step 3

4Step 4

5Step 5

Trap 1: Step 1: Achieve 3NF by removing transitive dependencies

This is incorrect because normalization must start from the lowest normal form. Removing transitive dependencies (3NF) before ensuring atomicity and removing partial dependencies is logically impossible and fails the prerequisite conditions.

Trap 2: Step 1: Achieve 2NF by removing partial dependencies

This is incorrect because second normal form (2NF) requires the table to already be in first normal form (1NF). Removing partial dependencies before achieving 1NF violates the normalization hierarchy.

Trap 3: Step 1: Achieve 1NF by removing repeating groups

This is incorrect because third normal form (3NF) requires the table to already be in second normal form (2NF). Removing transitive dependencies before addressing partial dependencies skips a necessary step and may lead to incorrect decomposition.

A
Step 1: Achieve 1NF by removing repeating groups and ensuring atomic values. Step 2: Achieve 2NF by removing partial dependencies. Step 3: Achieve 3NF by removing transitive dependencies.
This is the correct order because normalization proceeds sequentially from first normal form (1NF) to second normal form (2NF) to third normal form (3NF). Each step builds on the previous one: first address atomicity and repeating groups, then partial dependencies, then transitive dependencies.
B
Step 1: Achieve 3NF by removing transitive dependencies. Step 2: Achieve 2NF by removing partial dependencies. Step 3: Achieve 1NF by removing repeating groups.
Why wrong: This is incorrect because normalization must start from the lowest normal form. Removing transitive dependencies (3NF) before ensuring atomicity and removing partial dependencies is logically impossible and fails the prerequisite conditions.
C
Step 1: Achieve 2NF by removing partial dependencies. Step 2: Achieve 1NF by removing repeating groups. Step 3: Achieve 3NF by removing transitive dependencies.
Why wrong: This is incorrect because second normal form (2NF) requires the table to already be in first normal form (1NF). Removing partial dependencies before achieving 1NF violates the normalization hierarchy.
D
Step 1: Achieve 1NF by removing repeating groups. Step 2: Achieve 3NF by removing transitive dependencies. Step 3: Achieve 2NF by removing partial dependencies.
Why wrong: This is incorrect because third normal form (3NF) requires the table to already be in second normal form (2NF). Removing transitive dependencies before addressing partial dependencies skips a necessary step and may lead to incorrect decomposition.

Question 20mediumdrag order

Drag and drop the steps to implement a data classification policy in the correct order.

Drag steps to the numbered slots on the right, or tap a step then tap a slot.

Steps

Order

1Step 1

2Step 2

3Step 3

4Step 4

5Step 5

Trap 1: Train employees, define classification levels, assign ownership,…

Training should occur after the policy is implemented, not before, as users need to understand the existing labels and rules. Starting with training without defined levels or labels would be ineffective.

Trap 2: Label data, define classification levels, assign ownership, enforce…

Labeling data without first defining classification levels leads to inconsistent or incorrect labels. Levels must be defined to guide what labels to apply.

Trap 3: Define classification levels, label data, assign ownership, enforce…

Assigning ownership should come before labeling because owners are responsible for deciding the appropriate label for their data. If you label first, ownership may not align with labeling decisions.

A
Define classification levels, assign ownership, label data, enforce access control, train employees
This order is correct because you must first define the categories (levels), then assign responsibility (ownership), then apply labels to data, then restrict access based on labels, and finally educate users on the policy.
B
Train employees, define classification levels, assign ownership, label data, enforce access control
Why wrong: Training should occur after the policy is implemented, not before, as users need to understand the existing labels and rules. Starting with training without defined levels or labels would be ineffective.
C
Label data, define classification levels, assign ownership, enforce access control, train employees
Why wrong: Labeling data without first defining classification levels leads to inconsistent or incorrect labels. Levels must be defined to guide what labels to apply.
D
Define classification levels, label data, assign ownership, enforce access control, train employees
Why wrong: Assigning ownership should come before labeling because owners are responsible for deciding the appropriate label for their data. If you label first, ownership may not align with labeling decisions.