Short, precise answers to the machine-learning questions asked most often — each points back to the topic that explains it in depth.
Fundamentals
What is the difference between supervised, unsupervised and reinforcement learning?
What is overfitting and how do you prevent it?
Explain the bias–variance trade-off.
Why do we split data into train, validation and test sets?
What is cross-validation and when would you use GroupKFold or TimeSeriesSplit?
What is data leakage? Give two examples.
Algorithms
How does logistic regression work?
L1 vs L2 regularisation?
How does a decision tree choose splits, and why do trees overfit?
Random forest vs gradient boosting?
How do SVMs work? What is the kernel trick?
How does k-NN work and what are its weaknesses?
Why is Naive Bayes naive, and why does it still work?
How does k-means work and what are its assumptions?
What is PCA and why scale before it?
Evaluation
Precision vs recall — when do you favour each?
ROC-AUC vs PR-AUC?
Can R² be negative?
How do you handle imbalanced data?
What is model calibration and how do you check it?
How would you compare two models statistically?
Features & data
How do you handle missing values?
One-hot vs label (ordinal) encoding?
Which models need feature scaling?
What is the curse of dimensionality?
Production & responsible AI
What is concept drift vs data drift?
What is training–serving skew?
How do you explain a black-box model's predictions?
How do you measure fairness?
Beyond classic ML
When would you use deep learning instead of classic ML?
What is the relationship between logistic regression and a neural network?