Top 30 Data Science & Analytics Interview Questions and Answers

Question 1

When converting a large list of flat dictionaries to a pandas DataFrame, which method is the idiomatic one-shot approach for aligning keys and avoiding repeated DataFrame growth?

Accepted Answer

Pass the list of dicts directly to pd.DataFrame so the constructor aligns the dictionary keys in one operation

Answer

Convert the list of dicts into a list of lists first, then pass that to pd.DataFrame

Answer

Use DataFrame.from_dict with orient='index' and then transpose the result

Answer

Iterate through the list and build the DataFrame incrementally using pd.concat for each dictionary

Question 2

For a DataFrame with a string index, which call raises an error and why?

Accepted Answer

df.iloc['x'], because iloc requires integer positions not labels

Answer

df.loc[:, :], because full slices are disallowed

Answer

df.iloc[0], because position 0 is reserved

Answer

df.loc['x'], because loc cannot take a single label

Question 3

A pandas DataFrame has 5% missing values in customer_age, completely at random. When is dropna the better choice over fillna?

Accepted Answer

When generating a quick aggregate report and the remaining 95% of data is sufficient

Answer

When building a churn model where age is predictive and each row contains a labeled outcome

Answer

When the column uses nullable Int64 dtype to avoid coercion to float64

Answer

When you want to add an is_age_missing indicator column to preserve missingness signal

Question 4

Which query pattern correctly identifies customers who have never placed an order?

Accepted Answer

LEFT JOIN orders then filter WHERE orders.customer_id IS NULL

Answer

Filter with WHERE customer_id NOT IN (SELECT customer_id FROM orders)

Answer

LEFT JOIN orders then filter WHERE orders.customer_id = NULL

Answer

INNER JOIN customers to orders and select DISTINCT names

Question 5

Which approach correctly handles the full response lifecycle when fetching JSON from a REST API in production using Python's requests?

Accepted Answer

Verify the status with r.raise_for_status() before parsing with r.json()

Answer

Set a request timeout but skip status checks to parse faster

Answer

Use r.text and manually split the string to extract JSON fields

Answer

Pass query parameters in the request body and call r.json() directly

Question 6

Why is a violinplot generally preferred over a bar plot when exploring house prices across neighborhoods?

Accepted Answer

It preserves the full distribution, revealing features like bimodality and outliers that a bar plot's single statistic would obscure.

Answer

It is the default choice for non-technical audiences because it requires no explanation of medians or quartiles.

Answer

It avoids overplotting by displaying every individual data point in a beeswarm arrangement.

Answer

It summarizes each neighborhood with a single estimate and confidence interval for faster interpretation.

Question 7

During EDA, a heatmap reveals a strong positive correlation between daily ice cream sales and sunscreen sales. What is the most appropriate next step before making causal business recommendations?

Accepted Answer

Investigate potential confounders such as outdoor temperature and examine partial correlations before making causal recommendations.

Answer

Dismiss the correlation as irrelevant because correlation metrics cannot prove causation and therefore provide no actionable insight.

Answer

Launch a joint promotion because the heatmap indicates a causal link between the two product categories.

Answer

Immediately remove one variable from the dataset since two highly correlated features are always redundant in predictive models.

Question 8

A data scientist chooses Z-score standardization over Min-Max scaling before training a logistic regression model with gradient descent. What is the primary statistical reason for this choice?

Accepted Answer

Z-score standardization centers features at mean zero with unit variance, ensuring weight updates proceed at comparable rates during gradient descent.

Answer

Min-Max scaling bounds every feature to the same range, which prevents any single feature from dominating gradient descent weight updates.

Answer

Z-score standardization eliminates outliers entirely, producing a cleaner loss surface than Min-Max scaling would.

Answer

Logistic regression requires no scaling because it relies on rank ordering and is therefore scale-invariant like tree-based models.

Question 9

When a categorical feature has thousands of unique values, which approach avoids the dimensionality and memory issues of one-hot encoding while keeping predictive signal?

Accepted Answer

Target encoding, which replaces each category with the mean of the target variable for that category

Answer

Label encoding, which assigns a unique integer to each category so the feature remains a single column

Answer

Ordinal encoding, which sorts categories by frequency and replaces them with their rank

Answer

Dropping the feature to reduce model complexity and training time

Question 10

In fraud detection with 1% fraud prevalence, a model labels every transaction as legitimate and achieves 99% accuracy. What is the key lesson?

Accepted Answer

Accuracy can be misleading because it ignores the rare positive class entirely

Answer

Maximizing recall alone is the best strategy since catching fraud matters most

Answer

The model should be accepted because 99% accuracy exceeds most business benchmarks

Answer

High accuracy guarantees that a model has strong predictive power

Question 11

Which description best captures how K-Means iteratively forms its final clusters?

Accepted Answer

It initializes K centroids, then repeatedly assigns points to the nearest centroid and recomputes each centroid as the cluster mean until stable.

Answer

It minimizes straight-line Euclidean distance by iteratively shifting each centroid to the median of its assigned points.

Answer

It selects K random points as fixed centers and assigns every other point to the nearest center in one step.

Answer

It divides the data into K equally sized partitions and then computes the mean of each partition.

Question 12

You run an experiment and obtain p = 0.03. Which statement correctly interprets this p-value?

Accepted Answer

If the null hypothesis were true, there is a 3% probability of observing data at least this extreme.

Answer

There is a 97% probability that the alternative hypothesis is true.

Answer

There is a 3% chance that the null hypothesis is false.

Answer

The result has only a 3% probability of being due to random chance.

Question 13

Which hypothesis pair best tests whether a green signup button increases registrations compared to blue?

Accepted Answer

H0: The registration conversion rate for green equals that for blue; H1: The registration conversion rate for green is strictly greater than that for blue.

Answer

H0: There is no difference in registration conversion rate; H1: The registration conversion rates are not equal.

Answer

H0: The blue button has a higher registration conversion rate than the green button; H1: The green button has a higher registration conversion rate than the blue button.

Answer

H0: The green and blue buttons perform the same; H1: The green button performs better.

Question 14

Which approach best demonstrates a sound framework for measuring e-commerce user engagement?

Accepted Answer

Use DAU/MAU, adoption, retention, and stickiness tied to conversion and churn

Answer

Track total page views and raw session counts as primary success indicators

Answer

Collect qualitative feedback as the main quantifiable engagement signal

Answer

Monitor as many metrics as possible without ranking their business impact

Question 15

An e-commerce site sees search users convert at 5% and non-search users at 2%. What is the best next step to estimate the true incremental impact of search?

Accepted Answer

Match searchers to similar non-searchers using device, landing page, and prior sessions, then compare rates

Answer

Run a two-sample t-test on the conversion rates to confirm the difference is significant

Answer

Immediately design an A/B test that hides the search bar for the control group

Answer

Report the 150% lift as proof that search causes higher conversions

Question 16

A monthly subscriber cancels mid-cycle but retains access until month-end. Which approach reflects a rigorous churn definition for this user?

Accepted Answer

Waiting until the billing period ends or a reactivation window passes

Answer

Counting churn the instant the user clicks cancel

Answer

Flagging churn immediately when the next payment fails

Answer

Labeling churn after 30 days of inactivity while billing continues

Question 17

A checkout A/B test shows significantly higher conversion but slightly lower AOV. What is the strongest basis for a launch recommendation?

Accepted Answer

Estimate net revenue and ensure the AOV decline is within a pre-specified non-inferiority margin

Answer

Launch because conversion is up and automatically outweighs any decrease in basket size

Answer

Require AOV to also achieve statistical significance as a co-success metric

Answer

Adjust p-values for multiple comparisons and launch only if both metrics remain significant

Question 18

A marketing team debates predicting exact spend versus High/Medium/Low tiers. What most strongly determines whether regression or classification is the better framing?

Accepted Answer

Whether the downstream campaign action requires a ranked list and continuous segmentation or a hard gate into fixed segments.

Answer

The preference of business stakeholders for yes-or-no answers over probabilistic scores.

Answer

The algorithm that achieves the highest cross-validated R-squared or F1 score on historical data.

Answer

The degree of class imbalance in the historical customer base, since severe imbalance favors regression.

Question 19

Why is the correlation between profile completion and retention weak evidence that forcing completion will improve retention?

Accepted Answer

Engaged users self-select into completing profiles, confounding the relationship

Answer

Correlation coefficients cannot apply to user behavior data

Answer

Retention can never be measured accurately for new users

Answer

The sample of completed profiles is always too small to trust

Question 20

What is the most important addition to a revenue metric when evaluating a change that may harm long-term satisfaction?

Accepted Answer

Long-horizon guardrail metrics like retention plus a sufficiently long experiment

Answer

Measuring revenue more precisely with a larger sample

Answer

Shipping quickly to capture the revenue before users notice

Answer

Relying on stakeholder intuition about user happiness

Question 21

When validating a composite burnout proxy, why test whether it predicts voluntary attrition six months later?

Accepted Answer

It establishes predictive validity by showing the proxy correlates with a meaningful future outcome

Answer

It confirms the survey instrument has higher reliability than behavioral telemetry

Answer

It proves burnout causally drives attrition rather than merely correlating with it

Answer

It satisfies convergent validity by aligning the proxy with self-reported measures

Question 22

What justifies using a z-test for a population mean when the underlying data are heavily skewed?

Accepted Answer

The sampling distribution of the sample mean becomes approximately normal for large n

Answer

The sample mean converges to the population mean as sample size increases

Answer

The raw observations in large samples themselves become normally distributed

Answer

The population variance is known and the sample size exceeds thirty

Question 23

Which statement correctly describes how increasing model complexity affects bias and variance?

Accepted Answer

Bias decreases but variance increases because the fit depends more on the specific training sample.

Answer

Bias decreases while variance stays low until the model passes an overfitting threshold and variance spikes.

Answer

Bias decreases while variance stays unchanged, because variance depends only on training set size.

Answer

Both bias and variance decrease as the model fits the training data more closely.

Question 24

A medical model reports 99% accuracy on a disease with 1% prevalence. Which observation best shows why accuracy is misleading?

Accepted Answer

The model could predict every patient is healthy and still score 99%.

Answer

Precision and recall become identical when the positive class is rare.

Answer

The dataset is too small to yield a reliable accuracy estimate.

Answer

High accuracy proves the model has a low false positive rate.

Question 25

What happens when the learning rate in gradient descent is set too high?

Accepted Answer

Steps overshoot the minimum and the loss may oscillate or diverge

Answer

The gradient is computed in the wrong direction

Answer

The model converges to the global minimum guaranteed

Answer

Training always reaches the minimum faster with no downside

Question 26

In PCA on mean-centered data, what does the eigenvalue associated with a principal component directly represent?

Accepted Answer

The variance of the data projected onto that component's direction

Answer

The number of samples lying on that axis

Answer

The correlation between the original two features

Answer

The classification accuracy gained by keeping that component

Question 27

In a two-proportion conversion test, which change most directly explains why detecting a 2% relative lift requires roughly 390k users per variant instead of 63k for a 5% relative lift?

Accepted Answer

The absolute MDE shrinks from 0.5 to 0.2 percentage points, and sample size scales roughly with the inverse square of the absolute effect size.

Answer

Lowering the relative lift from 5% to 2% reduces statistical power from 80% to approximately 20% unless sample size is inflated.

Answer

The standard error of a proportion depends on p(1-p), which drops nonlinearly when the treatment rate approaches the baseline rate.

Answer

The baseline conversion rate of 10% creates a ceiling effect that exponentially increases variance as the relative lift decreases.

Question 28

Under what condition does MAP estimation give essentially the same result as MLE?

Accepted Answer

When the prior is uniform or the dataset is very large

Answer

When the dataset is very small

Answer

When the likelihood is non-convex

Answer

When the prior is very strong and informative

Question 29

You need to combine customers and transactions on customer_id to compute total spend per customer, including those with zero transactions. Which approach correctly uses pandas?

Accepted Answer

Use pd.merge with how='left' on customer_id, then group by customer, sum amounts, and fill NaN totals with zero

Answer

Use pd.concat to align the tables on customer_id, then group by customer and sum amounts

Answer

Use pd.merge with how='inner' on customer_id, then group by customer and sum amounts

Answer

Use pd.merge with how='right' on customer_id, then group by customer and sum amounts

Question 30

When squaring a million-element NumPy array, why is np.square(arr) orders of magnitude faster than a Python for-loop?

Accepted Answer

It delegates to pre-compiled C loops on contiguous memory, avoiding Python interpreter overhead per element.

Answer

It automatically distributes the computation across multiple CPU threads.

Answer

It removes explicit loops by expressing the entire operation in a single Python statement.

Answer

It uses Python list comprehensions that the interpreter optimizes for numerical arrays.