Principle:Scikit learn contrib Imbalanced learn Dataset Imbalancing
| Knowledge Sources | |
|---|---|
| Domains | Machine_Learning, Data_Preprocessing, Testing |
| Last Updated | 2026-02-09 03:00 GMT |
Overview
A utility technique for artificially introducing class imbalance into a balanced dataset to create controlled test scenarios for imbalanced learning methods.
Description
Dataset imbalancing is a testing and demonstration utility that takes a balanced dataset and reduces the number of samples in specified classes to create a controlled imbalanced distribution. This is useful for benchmarking resampling algorithms, demonstrating the effects of class imbalance, and generating reproducible test datasets.
Usage
Use this principle when creating toy or benchmark datasets to test resampling algorithms or demonstrate the impact of class imbalance on classifier performance.
Theoretical Basis
The imbalancing operation is a deterministic sub-sampling: for each target class, retain only the specified number of samples (randomly selected). This is equivalent to applying random under-sampling with specific target counts.