🧪 Random Undersampling

Random Undersampling#

Drops random majority-class rows to balance classes — simple, but can discard useful information.

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 Despite careful review, it may still contain mistakes, omissions, or out-of-date information.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#

Random undersampling balances an imbalanced dataset the simplest way possible: by randomly deleting majority-class rows until the classes are closer in size. It is the counterpart of oversampling, which adds minority rows rather than removing majority ones.

Why it’s used#

When the majority class dominates (fraud, churn), a classifier can drift toward always predicting it. Trimming the majority restores balance so the model gives the minority class real weight — and, as a bonus, trains faster on the smaller set.

How it works#

With 10,000 majority and 1,000 minority rows, draw 1,000 of the majority at random and keep all the minority, yielding a balanced 1,000 vs 1,000.

Trade-offs#

Advantages:

Simple and fast.
Balances the classes so the minority is not ignored.
Less data means quicker training.

Disadvantages:

Information loss — useful majority rows are thrown away.
Risk of underfitting from training on fewer points.
Poor when the minority class is tiny, since too much majority data is discarded.

Smarter alternatives#

Random oversampling or SMOTE — grow the minority instead of shrinking the majority.
Tomek links / Edited Nearest Neighbours — remove borderline or overlapping majority points rather than random ones.
Ensemble methods — combine undersampling with bagging/boosting, e.g. a Balanced Random Forest.

Example#

from collections import Counter
from imblearn.under_sampling import RandomUnderSampler

print("before:", Counter(y))
X_res, y_res = RandomUnderSampler(random_state=42).fit_resample(X, y)
print("after: ", Counter(y_res))

Theme: Imbalanced Learning & Resampling · All terminology

Hint

Mind map — connected ideas

Oversampling · SMOTE (Synthetic Minority Over-sampling Technique) · NearMiss (Distance-based Undersampling) · Cluster-based undersampling · Class Weighting · Subsampling

Hint

More in Imbalanced Learning & Resampling

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