EditedNearestNeighbors<T>

Models & Types · AiDotNet.Preprocessing.ImbalancedLearning

Implements Edited Nearest Neighbors (ENN) undersampling for handling imbalanced datasets.

Implements: IResamplingStrategy<T>

For Beginners

ENN uses a simple rule to decide which samples to remove:

“If a sample’s neighbors mostly disagree with its label, remove it.”

For example, if a majority class sample has 3 nearest neighbors:

  • 2 are minority class
  • 1 is majority class

Then the sample is “misclassified” by its neighbors and is removed.

This is like asking: “Would a K-NN classifier with K=3 correctly classify this sample?” If no, the sample is probably noisy or on the wrong side of the boundary.

Visual example:

Before: M M M m M m m m m
^
This M surrounded by m's would be removed

After: M M M m m m m m
Cleaner boundary

When to use:

  • Data cleaning before training
  • In combination with SMOTE (SMOTE + ENN = SMOTEENN)
  • When you want to remove ambiguous/noisy samples

References:

  • Wilson (1972). “Asymptotic Properties of Nearest Neighbor Rules Using Edited Data”

How It Works

ENN removes samples whose class label differs from the majority of their k nearest neighbors. This removes noisy and borderline samples from the majority class.

Constructors

Constructor Summary
EditedNearestNeighbors(int kNeighbors, ENNKind kind) Initializes a new instance of the EditedNearestNeighbors class.

Properties

Property Summary
string Name { get; } Gets the name of this undersampling 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 by removing samples misclassified by their neighbors.