Machine Learning Interview Questions and Answers (2025) | JaganInfo

Machine Learning is a branch of AI where systems learn patterns from data without being explicitly programmed.

Supervised, Unsupervised, Semi-supervised, and Reinforcement Learning.

Model trained on labeled data to predict outcomes for new, unseen inputs.

Model trained on unlabeled data to find patterns like clusters or associations.

An agent learns to make decisions by taking actions in an environment to maximize rewards.

When a model performs well on training data but poorly on new data due to memorizing noise instead of general patterns.

When a model is too simple to capture data patterns, resulting in poor performance on both training and test sets.

Spam detection, recommendation systems, fraud detection, speech recognition, medical diagnosis.

A mathematical representation of learned patterns that can make predictions on new inputs.

AI is the overall concept of intelligent machines; ML is a subset focused on learning from data.

A model splitting data into branches based on feature conditions to arrive at a decision or prediction.

An ensemble learning method combining multiple decision trees to improve prediction accuracy and control overfitting.

Balancing error due to bias (assumptions) and variance (sensitivity to data), aiming for optimum model complexity.

For binary or multi-class classification tasks using a logistic (sigmoid) function.

Classification predicts categories; regression predicts continuous values.

A technique to prevent overfitting by adding a penalty to the model complexity in the loss function (L1, L2).

Standardizing or normalizing input features to improve training stability and convergence speed.

Principal Component Analysis reduces dimensionality by transforming features into uncorrelated components capturing most variance.

Classifiers that find the optimal hyperplane separating classes with maximum margin.

An interconnected group of nodes simulating brain neuron behavior to learn data patterns.

An unsupervised algorithm partitioning data into k clusters by minimizing intra-cluster variance.

A method to evaluate model performance by training/testing on different data subsets.

An iterative optimization algorithm used to reduce loss by updating weights in opposite direction of gradients.

Settings like learning rate, batch size, epochs, chosen before training starts and not learned from data.

A table showing correct vs incorrect predictions for classification problems, used to calculate precision, recall, etc.

Precision: proportion of correct positive predictions; Recall: proportion of actual positives correctly predicted.

The harmonic mean of precision and recall, useful for imbalanced datasets.

A plot of true positive rate vs false positive rate to visualize classification performance trade-offs.

Combining multiple models to produce better predictive performance than a single model.

Extreme Gradient Boosting is an efficient, scalable implementation of gradient boosted decision trees, known for high performance in ML competitions.

Bagging trains models in parallel on random data subsets; boosting trains sequentially, correcting errors of prior models.

Combining multiple base model predictions via a meta-model to improve accuracy.

When feature space becomes high-dimensional, data becomes sparse, making models prone to overfitting and slow computation.

Vector representations of words capturing semantic meaning, useful in NLP tasks.

Includes Q-learning, Deep Q-Networks, and Policy Gradient methods where agents learn optimal actions through rewards.

Combines deep neural networks with reinforcement learning to make decisions in complex environments.

Using a model trained on one task as a starting point for a different but related task, reducing the required data and training time.

Batch learning trains on the whole dataset; online learning updates the model incrementally as new data arrives.

Understanding how a model arrives at its decisions, using techniques like SHAP, LIME for transparency and trust.

Identifying rare instances in data that differ significantly from the majority, e.g., fraud detection.

Adjusting the learning rate over epochs to improve convergence and avoid overshooting minima.

Models that learn the distribution of data to generate new, similar samples, e.g., GANs, VAEs.

Process of integrating a trained ML model into a production environment for real-world predictions.

A decentralized learning approach where models are trained across multiple devices without centralizing raw data.

Techniques where inputs are manipulated to fool models, highlighting security vulnerabilities.

Making predictions instantly as new data arrives, critical for applications like fraud detection.

When the statistical properties of target variables change over time, requiring model adaptation.

“Learning to learn” — developing models that learn new tasks quickly with minimal data.