Multiple-Strategies dung beetle optimizer and its applications in engineering optimization and bankruptcy prediction

Area of Science:

• Optimization algorithms
• Machine learning
• Engineering design

Background:

• Engineering optimization involves complex, constrained, multimodal problems.
• Machine learning hyperparameter tuning requires navigating vast combinatorial spaces.
• Existing algorithms may struggle with premature convergence and boundary constraints.

Purpose of the Study:

• Introduce the Multiple-Strategies Dung Beetle Optimizer (MSDBO).
• Enhance the Dung Beetle Optimizer (DBO) with adaptive strategies for exploration, exploitation, and convergence.
• Evaluate MSDBO's effectiveness in constrained engineering optimization and machine learning hyperparameter tuning.

Main Methods:

• Developed MSDBO integrating adaptive ball-rolling, optimal boundary control, and foraging enhancement strategies.
• Benchmarked MSDBO against DBO variants and metaheuristics on CEC2017 and CEC2022 benchmark suites.
• Applied MSDBO to constrained engineering design problems (speed reducer, pressure vessel, step-cone pulley) and corporate bankruptcy prediction using Kernel Extreme Learning Machine.

Main Results:

• MSDBO demonstrated statistically superior performance over 11 competing algorithms on benchmark suites.
• Ablation studies confirmed the critical contribution of each MSDBO strategy.
• MSDBO achieved lower-cost solutions for engineering design problems and outperformed others in bankruptcy prediction (82% accuracy, 85% specificity).

Conclusions:

• MSDBO is an efficient, robust, and broadly applicable optimizer.
• The algorithm effectively addresses challenges in constrained engineering optimization and data-driven prediction.
• MSDBO offers a valuable tool for complex optimization tasks in both engineering and machine learning domains.