Overfitting happens when a model learns training data too closely, including noise and unnecessary patterns. As a result, it performs very well on training data but poorly on unseen data. Techniques like cross-validation, regularization, pruning, and dropout are commonly used to reduce overfitting.

Underfitting occurs when a model is too simple to capture important patterns in the dataset. It performs poorly on both training and testing data. Increasing model complexity, improving feature engineering, or training for longer durations can help solve underfitting problems in machine learning systems.

Bias refers to errors caused by overly simple assumptions in the learning algorithm, leading to underfitting. Variance occurs when a model becomes highly sensitive to training data, causing overfitting. A good machine learning model maintains the right balance between bias and variance for better generalization.

Cross-validation is a model evaluation technique used to measure how well a machine learning model generalizes to unseen data. In k-fold cross-validation, the dataset is divided into multiple folds, where one fold is used for testing and the remaining folds are used for training.

Feature engineering is the process of creating, transforming, or selecting meaningful variables from raw data to improve model performance. It may include encoding categorical variables, scaling numerical values, extracting date features, or creating interaction variables that help machine learning models learn better patterns.

Regularization is a technique used to prevent overfitting by adding penalties to model complexity. L1 regularization, also called Lasso, reduces unnecessary features, while L2 regularization, known as Ridge, reduces large coefficient values. Regularization helps improve model generalization on unseen datasets.

Classification predicts discrete categories or labels, such as spam or not spam, while regression predicts continuous numerical values like house prices or sales revenue. Logistic Regression and Decision Trees are used for classification, whereas Linear Regression is commonly used for regression problems.

Logistic Regression is a supervised machine learning algorithm mainly used for classification problems. It predicts probabilities using the sigmoid function and classifies outputs into categories. It is widely used in fraud detection, customer churn prediction, email spam filtering, and medical diagnosis applications.

A Decision Tree is a supervised learning algorithm that splits data into branches based on conditions or feature values. It is easy to interpret and visualize. Decision Trees are commonly used for both classification and regression tasks but may suffer from overfitting without pruning.

Random Forest is an ensemble learning algorithm that combines multiple Decision Trees to improve prediction accuracy and reduce overfitting. Each tree is trained on random subsets of data and features. It performs well on large datasets and is widely used in industry applications.

XGBoost is an advanced gradient boosting algorithm designed for speed and high performance. It sequentially builds models by correcting errors from previous models. XGBoost includes regularization, parallel processing, and handling of missing values, making it one of the most commonly asked algorithms in interviews.

Gradient Descent is an optimization algorithm used to minimize the loss function in machine learning models. It updates model parameters iteratively by moving in the direction of the negative gradient. Variants include batch gradient descent, stochastic gradient descent, and mini-batch gradient descent.

A confusion matrix is a table used to evaluate classification models by comparing actual and predicted values. It contains True Positives, True Negatives, False Positives, and False Negatives. Metrics like precision, recall, accuracy, and F1-score are derived from the confusion matrix.

Precision measures how many predicted positive cases are actually correct, while recall measures how many actual positive cases are correctly identified by the model. Precision is important when false positives are costly, whereas recall is important when false negatives are critical.

F1-score is the harmonic mean of precision and recall. It provides a balanced evaluation metric, especially when datasets are imbalanced. A higher F1-score indicates better classification performance. It is widely used in fraud detection, medical diagnosis, and recommendation system evaluation.

ROC-AUC is a performance metric used for classification models. ROC represents the Receiver Operating Characteristic curve, while AUC measures the area under the curve. A higher AUC value indicates better model performance in distinguishing between positive and negative classes.

Feature selection is the process of identifying the most important variables for training machine learning models. It helps reduce dimensionality, improve model performance, decrease training time, and prevent overfitting. Common methods include filter methods, wrapper methods, and embedded techniques.

Dimensionality reduction reduces the number of input features while preserving important information. It improves computational efficiency and helps avoid overfitting. Principal Component Analysis (PCA) is one of the most popular dimensionality reduction techniques used in machine learning and data science.

Principal Component Analysis (PCA) is an unsupervised dimensionality reduction technique that transforms correlated features into fewer independent components. It helps reduce dataset complexity, improves visualization, and speeds up training while retaining most of the important variance from the original data.

K-Means is an unsupervised learning algorithm used for clustering similar data points into groups. It works by assigning points to the nearest cluster centroid and updating centroids iteratively. K-Means is commonly used in customer segmentation, recommendation systems, and market analysis.

Bagging trains multiple models independently and combines their predictions to reduce variance, while boosting trains models sequentially by correcting previous errors to reduce bias. Random Forest uses bagging, whereas algorithms like XGBoost, AdaBoost, and Gradient Boosting use boosting techniques.

Hyperparameter tuning is the process of selecting the best model configuration for improved performance. Parameters like learning rate, tree depth, and batch size are optimized using methods such as Grid Search, Random Search, or Bayesian Optimization to achieve better accuracy and generalization.

Data leakage occurs when information from the test dataset unintentionally influences the training process, leading to unrealistic model performance. It can happen during preprocessing or feature engineering. Proper train-test splitting and preprocessing pipelines help prevent leakage in machine learning workflows.

Imbalanced data occurs when one class has significantly more samples than another class. This can cause models to favor the majority class. Techniques like oversampling, undersampling, SMOTE, and class weighting are commonly used to handle imbalanced datasets effectively.

Ensemble learning combines predictions from multiple machine learning models to improve overall performance and stability. It reduces errors and increases accuracy compared to individual models. Popular ensemble methods include Random Forest, Gradient Boosting, XGBoost, AdaBoost, and stacking techniques.

The learning rate controls how much model parameters are updated during training. A very high learning rate may overshoot the optimal solution, while a very low rate slows training. Choosing the right learning rate is essential for stable and efficient model convergence.

Model deployment is the process of integrating a trained machine learning model into production environments where users or applications can access predictions. Common deployment tools include Flask, FastAPI, Docker, Kubernetes, and cloud platforms such as AWS, Azure, and Google Cloud.

Explainable AI refers to techniques that help humans understand how machine learning models make decisions. Methods like SHAP values, LIME, and feature importance improve transparency and trust. Explainability is especially important in healthcare, finance, and other high-risk applications.

Artificial Intelligence is the broader field of building intelligent systems. Machine Learning is a subset of AI where systems learn from data. Deep Learning is a subset of Machine Learning that uses neural networks with multiple layers to solve complex tasks like image recognition and NLP.