A Tri-Objective Method for Bi-Objective Feature Selection in Classification.

Area of Science:

• Machine Learning
• Data Science
• Computer Science

Background:

• Feature selection aims to minimize features while maximizing classification performance, a bi-objective challenge.
• Interactions between features necessitate exploring feature subset diversity beyond objective space performance.

Purpose of the Study:

• Propose a tri-objective method for bi-objective feature selection in classification.
• Incorporate feature subset diversity in the search space as a third objective.

Main Methods:

• Converted a bi-objective feature selection problem into a tri-objective one by adding a diversity objective.
• Introduced a novel initialization strategy and an offspring reproduction operator to enhance diversity and search ability.

Main Results:

• The proposed method effectively balances minimizing feature count, maximizing classification performance, and exploring diverse feature subsets.
• Experimental results on 20 real-world datasets demonstrate superior performance compared to existing methods.

Conclusions:

• The tri-objective approach enhances feature selection by considering feature diversity.
• The novel strategies improve the exploration of promising feature combinations for classification tasks.