MESPBO: Multi-Strategy-Enhanced Student Psychology-Based Optimization Algorithm for Global Optimization Problems and...

Area of Science:

• Machine Learning
• Engineering Design
• Optimization Algorithms

Background:

• Feature selection and continuous optimization are critical but challenging in machine learning and engineering.
• Existing Student Psychology-Based Optimization (SPBO) methods suffer from premature convergence and lack of population diversity.

Purpose of the Study:

• To propose a novel Multi-Strategy-Enhanced Student Psychology-Based Optimizer (MESPBO).
• To address the limitations of premature convergence and insufficient population diversity in optimization tasks.

Main Methods:

• Incorporated a hybrid heuristic initialization using Latin Hypercube Sampling and Gaussian perturbation.
• Implemented an adaptive dual-learning position update mechanism for exploration-exploitation balance.
• Utilized a hybrid opposition-based reflective boundary control for enhanced search stability.

Main Results:

• MESPBO outperformed 11 state-of-the-art algorithms on the CEC2017 benchmark suite across various dimensions.
• Achieved the highest average classification accuracy on 9 out of 10 real-world feature selection datasets, reaching 100% on some.
• Selected smaller feature subsets (typically 2-4 features) without compromising accuracy, showing a ~10% improvement over original SPBO.

Conclusions:

• MESPBO demonstrates superior convergence accuracy, robustness, and scalability in global optimization.
• The proposed method offers an effective trade-off between optimization accuracy and feature compactness.
• MESPBO exhibits strong adaptability and generalization for both global optimization and feature selection problems.