Data Scientist Interview Questions

It means the observed result is unlikely under the null hypothesis.

Interviewers expect you to separate p-value, confidence, sample size, effect size, and practical business importance.

They show a plausible range for an estimated value.

A good explanation avoids saying there is a fixed probability the true value is inside one computed interval; instead, describe repeated sampling behavior.

It happens when sampled data does not represent the target population.

Bias can come from selection, survivorship, non-response, instrumentation, or product exposure differences.

Use simple functions, clear names, tests, typing where useful, and small modules.

Interviewers look for readability, error handling, separation of concerns, and code that another engineer can change safely.

They produce values lazily without building a full list.

Generators are useful for streams, large files, and pipelines where memory efficiency matters.

Catch specific exceptions and keep recovery close to the failure.

Avoid broad except blocks unless re-raising or adding context. Good error handling makes failure modes explicit.

Define the objective, metric, baseline, data, and deployment constraints.

A strong answer checks whether ML is needed, creates a simple baseline, and validates data quality before model complexity.

A model learns training noise and fails to generalize.

Mention train-validation gaps, regularization, simpler models, more data, cross-validation, and leakage checks.

Match the metric to the business cost of errors.

Accuracy may fail for imbalance. Consider precision, recall, F1, ROC-AUC, PR-AUC, calibration, or ranking metrics.

Define hypothesis, primary metric, guardrails, sample size, and rollout plan.

Good answers include randomization, power, duration, segmentation, novelty effects, and decision criteria.

Metrics that protect user or business health while optimizing the primary goal.

Examples include latency, churn, complaint rate, revenue, or quality metrics that should not degrade.

Prioritize the decision framework and investigate segment-level tradeoffs.

Conflicts often reveal heterogeneous effects, measurement issues, or a product tradeoff that needs stakeholder alignment.

Activation is the user action that predicts long-term value.

It should be measurable, early in the journey, and validated against retention or business outcomes.

It compares groups that share a starting event or period.

Cohorts help separate user aging, seasonality, acquisition quality, and product changes.

Tie usage, outcome, and guardrail metrics to the feature goal.

Avoid measuring clicks alone. Look for behavior change, user value, quality, and unintended consequences.