SmoteTomekAugmenter<T>

Models & Types · AiDotNet.Augmentation.Tabular

Implements SMOTE-Tomek combination for imbalanced datasets.

Inheritance: AugmentationBase<T, Matrix<T>> → TabularAugmenterBase<T> → SmoteTomekAugmenter<T>

For Beginners

SMOTE-Tomek combines SMOTE oversampling with Tomek links cleaning. First, SMOTE generates synthetic minority samples, then Tomek links are removed to clean the class boundary and reduce noise.

How It Works

How it works:

  • Apply SMOTE to generate synthetic minority samples
  • Identify Tomek links (pairs of samples from different classes that are mutual nearest neighbors)
  • Remove the majority class samples from Tomek links

Benefits:

  • Balances classes while cleaning the decision boundary
  • Reduces noise introduced by SMOTE near the boundary
  • More robust than SMOTE alone

Reference: Batista et al., “A Study of the Behavior of Several Methods for Balancing Machine Learning Training Data” (2004)

Constructors

Constructor Summary
SmoteTomekAugmenter(int kNeighbors, double samplingRatio, double probability) Creates a new SMOTE-Tomek augmenter.

Properties

Property Summary
double SamplingRatio { get; } Gets the SMOTE sampling ratio.
int KNeighbors { get; } Gets the number of nearest neighbors used by SMOTE.

Methods

Method Summary
IDictionary<string, object> GetParameters()
ValueTuple<Matrix<T>, Vector<T>> ApplySmoteTomek(Matrix<T> data, Vector<T> labels, AugmentationContext<T> context) Applies SMOTE-Tomek to a labeled dataset.