AI and machine learning interview questions

Classification predicts a label or class.

Training data fits the model.

On imbalanced data, predicting the majority class always looks accurate but useless.

Scaling puts numeric features on similar ranges so distance-based and gradient methods behave well.

The target we want to predict for each example.

More data, simpler models, regularization, dropout, early stopping, and cross-validation help.

It summarizes how well scores rank positives above negatives across thresholds.

k-NN predicts from the labels of the k closest training points.

Gradients shrink through deep layers so early weights barely learn.

It counts true positives, false positives, true negatives, and false negatives.

Add trees that correct previous residuals step by step.

Never shuffle future into training.

Encoding categories with target means can leak if done before split.

An input signal the model uses to predict.

Counts of true/false positives and negatives.

Resample training data, class weights, or better metrics.

Estimating generalization by rotating folds.

Without non-linearities, stacked layers collapse to one linear map.

Use feature contributions, examples, and limits — not equations.

When data is tiny, interpretability is required, or a linear baseline already wins.