A Steering-Matrix-Based Multiobjective Evolutionary Algorithm for High-Dimensional Feature Selection.

Area of Science:

• Computational Intelligence
• Machine Learning
• Data Science

Background:

• Multiobjective evolutionary algorithms (MOEAs) show promise for feature selection (FS).
• High-dimensional feature selection presents challenges due to the curse of dimensionality.
• Existing MOEAs struggle with efficiency and effectiveness in high-dimensional spaces.

Purpose of the Study:

• To propose a novel steering-matrix-based multiobjective evolutionary algorithm (SM-MOEA) for high-dimensional feature selection.
• To enhance search efficiency and the quality of selected feature subsets.
• To overcome the limitations of existing algorithms in high-dimensional FS tasks.

Main Methods:

• Development of a steering matrix to guide population evolution, considering feature and individual importance.
• Introduction of dimensionality reduction and individual repairing operators based on the steering matrix.
• Design of an effective initialization and update strategy for the steering matrix.

Main Results:

• SM-MOEA demonstrated superior performance on 12 high-dimensional datasets (3000-13000 features).
• The algorithm achieved better results in terms of both the number and quality of selected features.
• Experimental results show significant improvements over state-of-the-art single-objective and MOEA algorithms for high-dimensional FS.

Conclusions:

• SM-MOEA offers an effective solution for high-dimensional feature selection.
• The steering matrix mechanism significantly enhances search efficiency and feature subset quality.
• The proposed algorithm represents a notable advancement in MOEAs for challenging FS problems.