🧪 Downsampling

Downsampling#

Reducing majority-class examples to balance a dataset.

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 Despite careful review, it may still contain mistakes, omissions, or out-of-date information.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

Downsampling (random undersampling) rebalances an imbalanced dataset the opposite way — by removing majority-class examples until the classes are closer to even. It keeps all the minority data and thins out the majority.

The risk#

Discarding majority examples can cause underfitting — the model loses informative cases and may miss the majority class’s general pattern. In extreme imbalance you may throw away the vast bulk of the data (99%+), damaging its representation.

When to use it#

Prefer downsampling when data is plentiful, since it is computationally efficient (less data to train on) and avoids the overfitting of duplication. As with upsampling, apply it to the training set only to avoid data leakage.

Theme: Imbalanced Learning & Resampling · All terminology

Hint

Mind map — connected ideas

Upsampling · Bootstrap · SMOTE (Synthetic Minority Over-sampling Technique) · Recall · Precision (a.k.a. Positive Predictive Value, PPV) · Model Stability

Hint

More in Imbalanced Learning & Resampling

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