DATA SCIENCE INTERVIEW PREP

5 Questions that actually get asked in DS interviews — with clean, no-fluff answers

Q: What is the difference between bias and variance?*
Bias = error from wrong assumptions (model too simple → underfitting).
Variance = error from sensitivity to training data (model too complex → overfitting).
Goal: minimize both → the "bias-variance tradeoff."

Q: Explain the difference between L1 and L2 regularization.*
L1 (Lasso): adds |weights| penalty → can shrink coefficients to exactly 0 → useful for feature selection.
L2 (Ridge): adds weights² penalty → shrinks coefficients smoothly, never to 0 → useful when all features matter a bit.

Q: How do you handle imbalanced datasets?*
- Resampling: SMOTE (oversample minority) or undersample majority
- Use metrics beyond accuracy: Precision, Recall, F1, AUC-ROC
- Class-weighted loss functions
- Anomaly detection framing if imbalance is extreme (e.g., fraud)

Q: What is the Central Limit Theorem and why does it matter in DS?*
CLT: the sampling distribution of the mean approaches a normal distribution as sample size grows, regardless of the population's original distribution.
Why it matters: lets us use normal-distribution-based tests (t-tests, confidence intervals) even on non-normal data, as long as sample size is large enough (usually n ≥ 30).

Q: You have a model with 95% training accuracy but 65% test accuracy. What's happening and how do you fix it?

Classic overfitting.
Fixes:
- Add regularization (L1/L2)
- Reduce model complexity / prune features
- Get more training data
- Use cross-validation
- Apply dropout (for neural nets) or early stopping

Which one tripped you up? Drop your answer in the comments before scrolling up

Follow for daily interview drops!

#DataScience #InterviewPreparation #MachineLearning #TechJobs #DataScienceInterview #CareerInTech #MLInterview #AIJobs #CodingInterview #TechCareers #PlacementPrep #DSA #DataAnalytics #JobSearch #TechStudents