💻Technical Questions
Q1Explain the bias-variance trade-off.
💡High bias = underfitting, high variance = overfitting. How regularization, ensemble methods, and cross-validation manage this trade-off.
Q2When would you use L1 regularization vs L2?
💡L1 (Lasso) produces sparse models by zeroing out features — use for feature selection. L2 (Ridge) shrinks all coefficients — use when all features are relevant.
Q3How do you handle class imbalance in a classification problem?
💡Resampling (SMOTE, undersampling), class weights, threshold tuning, choosing the right metric (F1, AUC-ROC over accuracy), ensemble methods (BalancedRandomForest).
Q4Explain the difference between bagging and boosting.
💡Bagging (Random Forest): parallel, reduces variance. Boosting (XGBoost, LightGBM): sequential, reduces bias. Trade-offs in speed and overfitting risk.
Q5What is a ROC-AUC curve and what does it measure?
💡True Positive Rate vs False Positive Rate at different thresholds. AUC = probability model ranks positive instance higher than negative. Range: 0.5 (random) to 1.0 (perfect).
Q6How does gradient descent work?
💡Minimize loss function by iteratively moving in direction of negative gradient. Variants: batch, stochastic, mini-batch. Learning rate tuning, momentum, Adam optimizer.
Q7Explain how a decision tree makes splits.
💡Information gain (Gini impurity or entropy). Best split maximizes class purity. Depth, min_samples control overfitting.
🧠Behavioral Questions
B1Tell me about a model you built that had real business impact.
💡Be specific: what business problem, model choice, feature engineering, performance metric, deployment, and quantified impact (revenue, cost, accuracy).
B2Describe a time your analysis led to a surprising or counterintuitive insight.
💡Show intellectual curiosity and rigorous validation. How did you verify the finding? How did you communicate it to skeptical stakeholders?
🎯Situational Questions
S1Your recommendation model accuracy dropped from 87% to 74% overnight. What do you do?
💡Data drift check first (feature distributions), pipeline integrity check, training data issues, model version issue. Systematic triage from data to model.
S2A stakeholder wants to deploy a model with 60% accuracy because 'it's better than random'. How do you respond?
💡Business impact analysis: what are FP/FN costs? What's the baseline (current process accuracy)? What does 60% mean for users? Propose minimum bar based on business context.
Must-Know Topics
- ✓Machine Learning (supervised, unsupervised, ensemble)
- ✓Statistics (hypothesis testing, regression, distributions)
- ✓Python (pandas, scikit-learn, matplotlib)
- ✓Feature Engineering
- ✓Model Evaluation (cross-validation, metrics)
- ✓SQL
- ✓A/B Testing
- ✓MLOps basics (model deployment, monitoring)
Common Interview Mistakes to Avoid
- ✗Optimizing for accuracy on imbalanced data
- ✗Data leakage in feature engineering (using future data)
- ✗Not establishing a baseline model before complex models
- ✗Skipping exploratory data analysis
- ✗Presenting model results without business interpretation
Frequently Asked Questions
What machine learning algorithms are most commonly tested in data scientist interviews?▼
Linear/logistic regression, decision trees, random forest, gradient boosting (XGBoost/LightGBM), k-means, PCA, and neural network basics. For deep learning roles: CNNs, RNNs, transformers, and attention mechanisms.
How important is statistics for data scientist interviews?▼
Very important. Probability distributions, hypothesis testing (t-test, chi-square, ANOVA), Bayesian vs frequentist thinking, confidence intervals, and central limit theorem are commonly tested. Think of stats as the foundation that validates your ML conclusions.
Do data scientists need to know SQL in interviews?▼
Yes — SQL is tested at most companies. Data scientists need to extract and manipulate data from databases. Window functions, CTEs, and complex joins are commonly tested. Python data manipulation (pandas) is also expected.
How do I prepare for machine learning system design questions?▼
Practice end-to-end ML system design: define the problem → collect data → feature engineering → model selection → training → evaluation → deployment → monitoring → A/B testing. Use examples like building a recommendation system, fraud detection, or search ranking.
What's the difference between data scientist and ML engineer roles?▼
Data scientists focus on experimentation, model building, and business insights. ML engineers focus on productionizing models, ML infrastructure, and scalable model serving. In India, many companies have hybrid DS/MLE roles, especially at startups.
Ready for your Data Scientist interview?
Make sure your resume gets you to the interview stage first. Get a free ATS score.
Score My Resume Free →