100 classic ML interview

What is supervised learning?

Learning from labeled data.

Which algorithm is typically used for regression?

Linear regression

What does overfitting mean?

The model fits noise in training data.

Which method is used to prevent overfitting?

Regularization

Which metric is most appropriate for regression evaluation?

MSE

Which technique is a dimensionality reduction method?

PCA

Which ML model assumes independence between features?

Naive Bayes

What is the purpose of cross-validation?

Evaluate generalization

Which learning paradigm uses trial-and-error interaction?

Reinforcement

Which objective does clustering solve?

Grouping similar samples

What does bias refer to in ML?

Error from assumptions in model

Which best describes gradient descent?

Search for stationary points via gradients

What is the cost function for linear regression?

Mean squared error

What does regularization do?

Encourages smaller weights

What is L1 regularization also called?

Lasso

Which algorithm lazily learns at prediction time?

KNN

What is an SVM margin?

Distance between hyperplanes and classifier boundary

Which kernel allows non-linear separation in SVM?

All of the above

What is entropy in decision trees?

Measure of impurity

Which technique improves prediction by combining many weak models?

Ensembling

What is bagging?

Sampling with replacement and training models

Random forests reduce...

Variance

Which gradient boosting algorithm is widely used?

XGBoost

Which clustering algorithm requires number of clusters k?

K-means

What is the elbow method used for?

Choosing k in clustering

What is underfitting?

Model performs poorly on training and test data

Which sampling technique balances imbalanced datasets?

SMOTE

Which metric is best for imbalanced binary classification?

ROC-AUC

What does logistic regression output?

Probability of class

Softmax is used for...

Multi-class classification

Which ML method learns boundaries maximizing margin?

SVM

Which algorithm is sensitive to feature scaling?

KNN

What is ROC curve plotting?

TPR vs FPR

Which technique reduces variance in prediction?

Bagging

Which optimization modifies weights per feature ranking?

Feature selection

Which learning method creates synthetic minority samples?

SMOTE

Which distance measure is used in KNN?

Euclidean distance

What is cross entropy used for?

Classification loss

What is early stopping?

Stopping training to reduce overfitting

What does AUC measure?

Area under ROC curve

Which algorithm uses impurity reduction?

Decision tree

What is grid search used for?

Hyperparameter tuning

Which boosting method reweights samples?

AdaBoost

What is the curse of dimensionality?

Distance metrics lose meaning in high dimension

Which reduces dimensionality while preserving variance?

PCA

What is a confusion matrix used for?

Classification evaluation

Which expresses Bayes theorem?

P(A|B)=P(B|A)P(A)/P(B)

Which distribution models binary outcomes?

Bernoulli

Which technique detects outliers?

Isolation forest

What does model variance indicate?

Sensitivity to training sample fluctuations

Which sampling is used in bagging?

Sampling with replacement

What technique merges multiple weak models sequentially?

Boosting

Which activation function outputs probability?

softmax

Which regularization shrinks weights to zero?

L1

Which type of error is false positive?

Type I

Which clustering method identifies density-based clusters?

DBSCAN

Which metric measures similarity in clustering?

Silhouette score

Which technique reduces overfitting in random forests?

Feature bagging

Which sampling is used in Monte Carlo simulation?

Random sampling

Which ML task predicts continuous values?

Regression

Which approach automates feature search?

AutoML

What is variance inflation factor used for?

Detect multicollinearity

Which term refers to prior + likelihood combination?

Posterior

What is bagging strongest at reducing?

Variance

Which technique automatically prunes trees?

Cost-complexity pruning

Which learning algorithm uses nearest neighbors for decision?

KNN

ROC curve compares classifier performance based on…

TPR and FPR

Which is an unsupervised method?

K-means clustering

Which reduces high-dimensional noise?

PCA

Which model creates axis-aligned splits?

Decision tree

What does boosting reduce?

Bias

Which method avoids exhaustive search?

Random search

Which model family supports interpretability easily?

Decision trees

Which metric handles probabilistic outputs?

Log loss

Which is a lazy learner?

KNN

Which model finds hyperplanes using margin optimization?

SVM

Which step is required in SVM to handle non-linearity?

Use kernels

Which model learns class conditional density?

Naive Bayes

Which approach uses gradient boosting with trees?

XGBoost

What is the decision boundary in logistic regression?

Linear hyperplane

Which model samples feature subsets per tree?

Random forest

Which ML evaluation metric is threshold independent?

ROC-AUC

Which algorithm maximizes posterior probability?

Bayesian classifier

Which ML concept balances bias and variance?

Regularization

Which technique visualizes classification performance?

Both

Which optimization adjusts model hyperparameters iteratively?

Bayesian optimization

Which improves weak classifiers sequentially using residuals?

Boosting

Which describes multicollinearity?

Features strongly correlated

Which method measures class imbalance impact?

All of the above

Which prevents exploding gradients?

Weight clipping

Which approach learns probabilistic boundaries?

Logistic regression

Which model learns hierarchical splits?

Decision trees

Which step makes ML models sensitive to data distribution?

Normalization

Which reduces dimensionality via eigen decomposition?

PCA

Which ML model uses maximum likelihood estimation?

Logistic regression

Which improves model generalization?

All of the above

Which method prevents data leakage?

Scaling after splitting

Which metric balances precision and recall?

F1 score

What does cross-entropy measure?

Distance between probability distributions

Which technique samples train data with replacement?

Bagging