DS-ARO: a multi-strategy improved artificial rabbits optimization algorithm for global optimization and corporate...

Area of Science:

• Optimization Algorithms
• Metaheuristics
• Computational Intelligence

Background:

• Existing Artificial Rabbits Optimization (ARO) algorithms struggle with high-dimensional, multimodal problems due to poor convergence guidance, exploration-exploitation balance, and adaptive control.
• Addressing these limitations is crucial for advancing complex optimization tasks in various scientific domains.

Purpose of the Study:

• To propose a novel Dual-Strategy Enhanced Artificial Rabbits Optimization (DS-ARO) algorithm to overcome the limitations of the standard ARO.
• To enhance convergence, population diversity, search reliability, and adaptive control in optimization processes.

Main Methods:

• Integration of three mechanisms: a convergence-diversity balanced mutation strategy, an adaptive elite-guided search, and a success-rate-based strategy selection.
• Comprehensive evaluation using CEC2017 and CEC2022 benchmark test suites against state-of-the-art metaheuristic algorithms.
• Application to hyperparameter optimization of K-Nearest Neighbors (KNN) for corporate bankruptcy prediction on the Wieslaw financial dataset.

Main Results:

• DS-ARO demonstrated superior optimization accuracy, faster convergence, and enhanced robustness compared to competing algorithms across benchmark test suites.
• Statistical analyses (Friedman, Wilcoxon) confirmed the significance of performance improvements.
• The DS-ARO-KNN model achieved higher accuracy, precision, recall, and F1-score in bankruptcy prediction, outperforming traditional KNN and other machine learning models.

Conclusions:

• The proposed DS-ARO algorithm effectively addresses the limitations of the standard ARO, offering improved performance for complex optimization problems.
• DS-ARO shows significant potential in practical applications, such as financial risk assessment through improved hyperparameter tuning for predictive models.