BorderlineSMOTE<T>

Models & Types · AiDotNet.Preprocessing.ImbalancedLearning

Implements Borderline-SMOTE for handling imbalanced datasets.

Inheritance: OversamplingBase<T> → BorderlineSMOTE<T>

For Beginners

Regular SMOTE creates synthetic samples uniformly among all minority samples. Borderline-SMOTE is smarter - it identifies which minority samples are “borderline” (near majority class samples) and only creates synthetics from those.

How it works:

  1. For each minority sample, count its majority neighbors
  2. Classify each minority sample as:
  • SAFE: Mostly minority neighbors (not borderline, skip)
  • DANGER: Mix of minority and majority neighbors (borderline, use for synthesis)
  • NOISE: Mostly majority neighbors (outlier, skip)
  1. Only generate synthetic samples from DANGER samples

Why this is better:

  • Creates samples where they’re needed most (at the boundary)
  • Avoids wasting effort on samples deep in minority territory
  • Reduces noise by not synthesizing from outliers

References:

  • Han et al. (2005). “Borderline-SMOTE: A New Over-Sampling Method in Imbalanced Data Sets Learning”

How It Works

Borderline-SMOTE focuses on minority samples that are near the decision boundary (borderline samples), as these are the most informative for classification.

Constructors

Constructor Summary
BorderlineSMOTE(double samplingStrategy, int kNeighbors, int mNeighbors, BorderlineSMOTEKind kind, int? seed) Initializes a new instance of the BorderlineSMOTE class.

Properties

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