Interview prep

Machine Learning Interview Questions

ML interviews split into modelling fundamentals and production concerns — data quality, drift and serving.

Questions and answers

01Explain the bias-variance trade-off.

High bias underfits and misses real structure; high variance overfits noise. Regularisation, more data and simpler models move you along that curve.

02How do you handle class imbalance?

Resampling, class weighting, threshold tuning and choosing metrics such as PR-AUC or recall at fixed precision instead of raw accuracy.

03What is data leakage and how do you catch it?

Information from the target or the future leaking into features. Suspiciously high validation scores, and a time-based split that collapses them, is the usual tell.

04How do you monitor a model in production?

Track input distribution drift, prediction distribution, latency and downstream business metrics, with a labelled holdout stream for delayed ground truth.

05When is a simpler model the better choice?

When interpretability, latency or maintenance cost matter more than the last few points of accuracy — which is most of the time in production.

How to answer these well

Interviewers are not grading recall — they are checking whether you've hit the failure mode the question describes. Anchor every answer to something you actually shipped or debugged with Machine Learning.

Say the trade-off out loud. "I'd use X here, but it costs Y under Z conditions" scores higher than a clean textbook definition every time.

Roles asking for Machine Learning

Related prep