A Cooperative Coevolutionary Approach to Discretization-Based Feature Selection for High-Dimensional Data

Area of Science:

• Machine Learning
• Data Mining
• Computational Intelligence

Background:

• Discretization-based feature selection methods integrate feature selection and discretization for high-dimensional data.
• Current methods often overlook feature interactions, leading to information loss.

Purpose of the Study:

• To propose a cooperative coevolutionary algorithm that addresses the limitations of existing methods.
• To simultaneously search for feature subsets with and without entropy-based cut-points.

Main Methods:

• A cooperative coevolutionary algorithm combining genetic algorithm (GA) and particle swarm optimization (PSO).
• GA with a ranking mechanism controls mutation and crossover for features with cut-points.
• Binary-coded PSO updates indices for features without cut-points.

Main Results:

• The proposed algorithm was tested on 10 real-world datasets.
• Demonstrated superior classification accuracy compared to state-of-the-art competitors.
• Effectively integrated feature selection and discretization by considering feature interactions.

Conclusions:

• The cooperative coevolutionary algorithm enhances feature selection for high-dimensional data.
• Simultaneous optimization of features and cut-points improves classification performance.
• The method effectively mitigates information loss by considering feature interactions.