SMOTE

Gold definitionUpdated Apr 2, 2026

Definition

SMOTE (Synthetic Minority Over-sampling Technique) is a data augmentation method that addresses class imbalance by generating synthetic samples for the minority class. It improves model performance and generalization, particularly for classifiers struggling with skewed data distributions.

At a glance

Executive summary

SMOTE is a technique used in machine learning to balance datasets where one class is much smaller than others. It creates new, synthetic examples of the rare class, helping models learn better and make more accurate predictions, especially for important but infrequent events. Its application can also influence the stability of model explanations.

TL;DR

SMOTE creates synthetic data points for rare classes to help AI models learn better from imbalanced datasets, improving prediction accuracy and influencing explanation consistency.

Key points

Generates synthetic minority class samples by interpolating between existing ones.
Solves the problem of biased models and poor minority class performance due to class imbalance.
Used by researchers and ML engineers in NLP (e.g., hate speech detection), finance (e.g., fraud), and healthcare (e.g., disease diagnosis).
Unlike simple oversampling (duplication), SMOTE creates novel samples, reducing overfitting.
Research trends include integrating SMOTE with deep learning, exploring its impact on model explainability, and developing adaptive variants.

Use cases

Fraud Detection: Identifying rare fraudulent transactions in vast datasets of legitimate ones.
Medical Diagnosis: Detecting rare diseases where positive cases are scarce compared to negative ones.
Hate Speech Detection: Improving the identification of minority class hate speech instances in large, imbalanced text datasets.
Anomaly Detection: Pinpointing unusual system behaviors or defects in manufacturing processes.
Credit Risk Assessment: Accurately identifying rare instances of loan defaults.

Also known as

Borderline-SMOTE, ADASYN, Safe-SMOTE, K-SMOTE, SMOTE-NC, SMOTE-N, Geometric SMOTE