Multivariate Discretization Based on Evolutionary Cut Points Selection for Classification.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Discretization is a key data preprocessing step for enhancing data understanding and enabling specific machine learning algorithms.
• Handling multivariate classification problems with high attribute interactions presents significant challenges.

Purpose of the Study:

• To propose an evolutionary algorithm for optimal data discretization.
• To address challenges in multivariate classification by selecting optimal cut points for numerical attributes.

Main Methods:

• Utilizing evolutionary algorithms to select a subset of cut points for discretization.
• Employing a wrapper fitness function to evaluate discretization schemes.
• Incorporating a reduction mechanism for managing large datasets in a multivariate context.

Main Results:

• The proposed algorithm demonstrated superior performance compared to state-of-the-art discretizers across 45 real-world datasets.
• Achieved competitive accuracy for classifiers like C4.5, Naive Bayes, PART, and PrUning and BuiLding Integrated in Classification.
• Generated significantly simpler discretization schemes.

Conclusions:

• Evolutionary algorithms offer an effective approach for multivariate data discretization.
• The method provides a robust solution for improving classification accuracy and model interpretability.
• This technique is particularly valuable for complex datasets where attribute interactions are prevalent.