Textbook of AI

Abstract. Introduces the Synthetic Minority Over-sampling Technique (SMOTE), the standard method for handling class imbalance in tabular classification. For each minority example, SMOTE selects a $k$-nearest minority neighbour and synthesises a new example along the line segment between them, sampling without exact duplication. SMOTE substantially outperforms naive minority oversampling and reweighting on imbalanced benchmarks. Two decades on, SMOTE and its variants (Borderline-SMOTE, ADASYN) remain the default class-imbalance baseline in imbalanced-learn.