NearMissUnderSampler<T>
Models & Types · AiDotNet.Augmentation.Tabular.Undersampling
Implements NearMiss undersampling for imbalanced datasets.
Implements: IUnderSampler<T>
For Beginners
NearMiss is a smarter undersampling method that selects majority class samples based on their distance to minority samples. Unlike random undersampling, it considers the spatial relationships between classes.
How It Works
NearMiss Variants:
- NearMiss-1: Selects majority samples whose average distance to k nearest
minority samples is smallest. These are majority samples close to the minority class.
- NearMiss-2: Selects majority samples whose average distance to k farthest
minority samples is smallest. These are majority samples close to all minority samples.
- NearMiss-3: For each minority sample, keeps the k nearest majority samples.
This surrounds each minority sample with selected majority samples.
When to use:
- NearMiss-1: When you want to keep majority samples near the decision boundary
- NearMiss-2: When minority class has outliers you want to handle
- NearMiss-3: When you want to preserve local structure around minority samples
Reference: Zhang & Mani, “kNN Approach to Unbalanced Data Distributions” (2003)
Constructors
| Constructor | Summary |
|---|---|
NearMissUnderSampler(NearMissVersion<T> version, int nNeighbors, double samplingRatio) |
Creates a new NearMiss undersampler. |
Properties
| Property | Summary |
|---|---|
NearMissVersion<T> Version { get; } |
Gets the NearMiss version to use. |
double SamplingRatio { get; } |
Gets the target ratio between minority and majority samples after undersampling. |
int NNeighbors { get; } |
Gets the number of nearest neighbors to consider. |
Methods
| Method | Summary |
|---|---|
ValueTuple<Matrix<T>, Vector<T>> Undersample(Matrix<T> data, Vector<T> labels, T minorityLabel) |
Performs NearMiss undersampling on the majority class. |