Interview questions covering machine learning fundamentals, from model selection to evaluation.
What's the difference between supervised and unsupervised learning?
Supervised learning trains on labeled data — you know the correct answer for each example. Unsupervised learning finds patterns in data without labeled outcomes, like clustering similar customers without predefined categories.
How do you know if a model is overfitting?
Strong performance on training data but significantly worse performance on validation or test data is the classic sign. Cross-validation helps catch this before deployment, when it's much cheaper to fix.
Explain precision and recall, and when each matters more.
Precision measures how many predicted positives were actually correct; recall measures how many actual positives were caught. In fraud detection, recall often matters more (missing fraud is costly). In a spam filter, precision often matters more (flagging real email as spam is annoying).
What's the bias-variance tradeoff?
High bias means a model is too simple and underfits; high variance means it's too complex and overfits, capturing noise as if it were signal. Good models balance the two — the right level of complexity for the actual amount and quality of data available.
How would you handle an imbalanced dataset?
Options include resampling (oversampling the minority class or undersampling the majority), using evaluation metrics beyond accuracy (like F1 score), or adjusting the model's decision threshold — the right approach depends on the specific cost of false positives versus false negatives.
Go deeper
These questions are just the surface
Our Machine Learning Track program builds the hands-on experience behind these answers — not just memorized responses.
