'Overfitting' in machine learning means:

Overfitting means high variance and low bias. The model memorizes training data rather than learning generalizable patterns. Overfitting occurs when training accuracy >> test accuracy. It can be prevented by Regularization (L1/Lasso adds |weights|, L2/Ridge adds weights² to loss function, penalizing complexity), Cross-validation (k-fold — evaluates on multiple train/test splits), Early stopping (stop training when validation loss increases), Dropout (neural networks — randomly deactivate neurons during training), Pruning (decision trees — remove low-information branches), More training data.  Underfitting means high bias and low variance resulting in model being too simple.