Binary Restructuring Particle Swarm Optimization and Its Application.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Particle Swarm Optimization (PSO) is a metaheuristic algorithm primarily for continuous optimization.
• Existing binary PSO variants often rely on transfer functions for discrete problems.
• There is a need for efficient binary optimization algorithms with robust exploration capabilities.

Purpose of the Study:

• To adapt the Restructuring Particle Swarm Optimization (RPSO) algorithm for discrete optimization problems.
• To introduce a novel binary variant, the binary Restructuring Particle Swarm Optimization (BRPSO) algorithm.
• To evaluate BRPSO's performance in feature selection tasks.

Main Methods:

• Developed the binary Restructuring Particle Swarm Optimization (BRPSO) algorithm, a novel adaptation of RPSO.
• BRPSO utilizes a unique particle updating mechanism based on position formula comparisons and random numbers, eschewing transfer functions.
• Incorporated a novel perturbation term into the position updating formula for enhanced exploration.

Main Results:

• BRPSO demonstrated competitive performance against four peer algorithms in feature selection experiments.
• The algorithm achieved high classification accuracy.
• BRPSO effectively reduced the number of selected features, indicating efficiency.

Conclusions:

• BRPSO is a viable and effective algorithm for discrete optimization problems, particularly feature selection.
• The algorithm's parameter efficiency and strong early-stage exploration capability are notable advantages.
• BRPSO offers a competitive alternative to existing binary metaheuristic algorithms.