Q1
Walk me through how you would approach handling class imbalance in a binary classification problem for fraud detection. What techniques have you used in production, and how did you evaluate their effectiveness?
Why they ask this:* Evaluates understanding of real-world ML challenges, knowledge of imbalance-handling methods (SMOTE, class weights, threshold tuning), and ability to select and validate appropriate solutions for data industry problems.
Q2
Explain the trade-offs between using a simpler model (like logistic regression) versus a complex ensemble model (like XGBoost) in a production recommendation system. How would you decide which to deploy?
Why they ask this:* Tests knowledge of model complexity, interpretability, inference latency, maintenance costs, and business considerations—critical for mid-level engineers making architecture decisions in data-driven companies.
Q3
Describe your experience with feature engineering in a real project. What features did you create, how did you validate their importance, and what tools or techniques did you use to manage feature pipelines?
Why they ask this:* Assesses practical experience with feature engineering (often 70% of ML work), understanding of statistical validation methods, and familiarity with MLOps tools needed for scalable data pipelines.
Q4
How do you approach debugging a machine learning model that performs well on validation data but poorly in production? Walk through your diagnostic process.