Edited Nearest Neighbor

[EmilHvitfeldt]

Feature Request

Add support for Edited Nearest Neighbor (ENN) under-sampling.

Description

Edited Nearest Neighbor is an under-sampling technique that removes samples whose class label differs from the majority of their k-nearest neighbors. This effectively removes noisy and borderline samples.

Algorithm

1. For each sample in the dataset:
   • Find its k-nearest neighbors
   • If the majority of neighbors belong to a different class, mark the sample for removal
2. Remove all marked samples

This is a cleaning method that removes misclassified or ambiguous samples, resulting in smoother decision boundaries.

Key Properties

• Removes noisy samples that are likely misclassified
• Cleans decision boundaries
• Can be applied to majority class only, or to all classes
• Often used as a preprocessing step before other methods

Relationship to Other Methods

• Extended by: Repeated Edited Nearest Neighbor (Repeated Edited Nearest Neighbor #173)
• Extended by: AllKNN (AllKNN #174)
• Related: Neighborhood Cleaning Rule (Neighborhood Cleaning Rule #116)

References

• Wilson, D. L. (1972). Asymptotic properties of nearest neighbor rules using edited data. IEEE Transactions on Systems, Man, and Cybernetics, (3), 408-421.
• https://imbalanced-learn.org/stable/references/generated/imblearn.under_sampling.EditedNearestNeighbours.html

[EmilHvitfeldt]

Related issues for ENN variants:

• Neighborhood Cleaning Rule #116 - Neighborhood Cleaning Rule
• Repeated Edited Nearest Neighbor #173 - Repeated Edited Nearest Neighbor
• AllKNN #174 - AllKNN

These are all variants/extensions of Edited Nearest Neighbor and could potentially share implementation infrastructure.