Evolutionary undersampling for classification with imbalanced datasets: proposals and taxonomy.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Learning with imbalanced data presents a significant challenge in machine learning.
• Existing solutions involve data modification or preprocessing techniques like undersampling and oversampling.
• Undersampling, a prototype selection method, aims to balance datasets for higher classification accuracy without majority class bias.

Purpose of the Study:

• To introduce and evaluate a novel set of methods termed evolutionary undersampling.
• To achieve a favorable trade-off between class distribution balance and classification performance.
• To compare evolutionary undersampling against state-of-the-art undersampling techniques.

Main Methods:

• Development of evolutionary undersampling algorithms incorporating problem-specific fitness functions.
• Creation of a taxonomy to categorize different evolutionary undersampling approaches.
• Comparative analysis using nonparametric statistical procedures to validate results.

Main Results:

• Evolutionary undersampling demonstrates superior performance compared to non-evolutionary models.
• The advantage of evolutionary undersampling becomes more pronounced as the degree of data imbalance increases.
• The proposed methods offer a robust solution for handling imbalanced datasets in machine learning.

Conclusions:

• Evolutionary undersampling is a promising technique for addressing the challenge of imbalanced data.
• The developed methods provide an effective strategy for improving classification performance in imbalanced scenarios.
• Further research into evolutionary algorithms for data preprocessing is warranted.