Data Scientist Machine Learning Interview Questions

Machine learning asked of a Data Scientist rarely stops at which model you would use. Across these 437 questions from Meta, Amazon, Google, Capital One and TikTok, the follow-up is where the round is decided: why that loss, what your metric does when the positive class is rare, how you would separate a genuine lift from a seasonal one, whether the feature you just proposed will actually exist at scoring time. Bias and variance, leakage, calibration, regularisation and the honest evaluation of a churn or fraud model recur throughout, alongside case-style prompts from consulting and credit teams where the first job is turning a business decision into a target variable. A quant strand runs through the slice as well, filed here from desks such as Two Sigma, where the modelling round doubles as an exercise in expected value, market making and correlation bounds. By rating this is the hardest Data Science topic on the site, 166 hard against 217 medium, and it arrives early rather than late: 240 came from a technical screen. 403 have a written solution, and none open a console, because the reasoning is the answer.