🧪 NearMiss (Distance-based Undersampling)

NearMiss (Distance-based Undersampling)#

Undersamples the majority class by keeping the majority points nearest to the minority, chosen by distance heuristics.

Important

✨ AI-generated content. This page was written with the assistance of an AI language model and is provided as a learning aid. Despite careful review, it may still contain mistakes, omissions, or out-of-date information. Whether you are new to the topic, a team lead, or a senior practitioner, treat it as a starting point rather than an authoritative reference: read it critically and independently verify anything you act on (code, commands, figures, and factual claims) against official documentation and primary sources before relying on it.

What it is#

NearMiss is a family of undersampling methods that shrink the majority class intelligently. Instead of dropping majority samples at random, NearMiss chooses which ones to keep based on their distance to minority samples, retaining the informative, hard-to-classify majority points near the boundary and discarding the “easy” ones far away.

The three versions#

NearMiss-1 — keep the majority samples whose average distance to their :math:`k` nearest minority samples is smallest (closest to the minority class).
NearMiss-2 — keep the majority samples whose average distance to their :math:`k` farthest minority samples is smallest.
NearMiss-3 — for each minority sample, keep a fixed number of its nearest majority samples, guaranteeing every minority point is surrounded.

Trade-offs#

Advantages:

Preserves boundary-defining points instead of throwing data away blindly.
Reduces majority-class bias and can generalise better than naive undersampling.

Disadvantages:

Distance computations make it more expensive than random undersampling.
Pruning “easy” samples too hard can overfit the difficult regions.
Results are sensitive to the choice of \(k\).

Example#

from collections import Counter
from imblearn.under_sampling import NearMiss

print("before:", Counter(y))
X_res, y_res = NearMiss(version=1).fit_resample(X, y)   # version 1, 2 or 3
print("after: ", Counter(y_res))

Theme: Imbalanced Learning & Resampling · All terminology

Hint

Mind map — connected ideas

Random Undersampling · Cluster-based undersampling · Subsampling · SMOTE (Synthetic Minority Over-sampling Technique) · Oversampling

Hint

More in Imbalanced Learning & Resampling

Class Weighting · Cluster-based undersampling · Downsampling · Oversampling · Random Undersampling · SMOTE (Synthetic Minority Over-sampling Technique) · Subsampling · Upsampling