AI and machine learning interview questions

Leakage is any path that lets information from the evaluation set, or from the future, into training.

Use chronological cuts so training always precedes validation and test.

A stratified split keeps class proportions similar in train, val, and test.

ROC-AUC is the probability that a random positive scores higher than a random negative.

Precision-recall AUC focuses on the positive class and reacts to prevalence.

It counts true positives, false positives, true negatives, and false negatives at a chosen threshold.

Log loss, or binary cross-entropy, penalizes confident wrong probabilities heavily.

R-squared is the fraction of variance in y explained by the model.

Multicollinearity means features move together so the design matrix is nearly singular.

The closed form needs a matrix inverse or solve on p features and can be costly or unstable for huge p.

Heteroscedasticity means residual variance changes with x or with the predicted value.

It minimizes log loss, which is the negative log likelihood of a Bernoulli model.

One-vs-rest fits a binary model per class.

5 only matches equal costs and a well-calibrated model with balanced classes.

It struggles with strongly nonlinear decision surfaces unless you add interactions or basis functions.

Pruning removes branches that add little validated gain so the tree does not memorize noise.

A feature with many levels can create many pure-looking splits by chance.

ID3 uses information gain and likes categorical splits.

Yes, order-preserving transforms such as log or min-max usually leave the split sequence unchanged for a single tree.

Each tree is trained on a bootstrap sample, so some rows are left out of that tree.