AI and machine learning interview questions

Classification votes among the K neighbors.

In high dimensions, distances concentrate and nearest neighbors stop being truly near.

Closer neighbors can vote more than far ones, which softens the effect of a clumsy K.

Silhouette compares how close a point is to its own cluster versus the next nearest cluster.

DBSCAN grows clusters from dense neighborhoods and labels sparse points as noise.

Inertia always falls as K grows, even when extra clusters are meaningless.

Use internal scores such as silhouette or Davies-Bouldin, plus visual checks on a 2D projection.

Look at the explained-variance ratio and keep components until a target such as 90 or 95 percent is reached.

PCA is better when many correlated numeric features share a low-dimensional subspace and you mainly need compression.

Trees split on original features and can ignore irrelevant axes without rotation.

Target encoding replaces a category with a statistic of y, such as the mean label.

One-hot becomes huge, so people use target encoding, hashing, embeddings, or frequency encoding.

Interactions help linear and logistic models capture AND-style effects they cannot learn from main effects alone.

If missingness itself predicts the target, mean imputation hides that signal.

SMOTE builds synthetic minority points by interpolating between a minority row and its neighbors.

Class weights leave the original rows intact and only change the loss.

It plots precision against recall as you move the threshold, focusing on the rare class.

Threshold tuning is often the first and cheapest lever when probabilities are usable.

ColumnTransformer applies different transformers to different column subsets in one step.

FeatureUnion runs transformers on the same data and concatenates their outputs.