SMOTETomek<T>

Models & Types · AiDotNet.Preprocessing.ImbalancedLearning

Implements SMOTE combined with Tomek Links (SMOTETomek) for handling imbalanced datasets.

Implements: IResamplingStrategy<T>

For Beginners

This is a two-step process:

Step 1 (SMOTE): Create synthetic minority samples

  • Increases minority class size
  • Fills in the minority class region

Step 2 (Tomek Links): Remove specific boundary pairs

  • Only removes samples that form Tomek links
  • More targeted than ENN
  • Preserves more data

Comparison with SMOTEENN:

  • SMOTETomek: Removes fewer samples, more conservative
  • SMOTEENN: Removes more samples, more aggressive cleaning

Visual example:

Original: M M M M M M M M M m m
After SMOTE: M M M M M M M M M m m m m m m m
^ ^
Tomek link pair
After Tomek: M M M M M M M M m m m m m m m
^
Only the majority sample of the pair removed

When to use:

  • When you want to balance but preserve more data
  • When ENN is too aggressive for your dataset
  • As a middle ground between pure SMOTE and SMOTEENN

References:

  • Batista et al. (2004). “A Study of the Behavior of Several Methods for Balancing

Machine Learning Training Data”

How It Works

SMOTETomek first applies SMOTE to oversample the minority class, then applies Tomek Links removal to clean up the decision boundary. This is less aggressive than SMOTEENN.

Constructors

Constructor Summary
SMOTETomek(double samplingStrategy, int kNeighbors, int? seed) Initializes a new instance of the SMOTETomek class.

Properties

Property Summary
string Name { get; } Gets the name of this resampling strategy.

Methods

Method Summary
ResamplingStatistics<T> GetStatistics() Gets statistics about the last resampling operation.
ValueTuple<Matrix<T>, Vector<T>> Resample(Matrix<T> x, Vector<T> y) Resamples the dataset using SMOTE followed by Tomek Links removal.