Abstract
Crayfish Optimization Algorithm (COA) suffers from degradation of diversity, insufficient exploratory capability, a propensity to become caught in local optima, and an imprecise search engine for optimization. To address these issues, the current research introduces a hybrid strategy enhanced crayfish optimization algorithm (MSCOA). Initially, a chaotic inverse exploration initialization method is utilized to establish the population's position with high diversity, significantly enhancing the global exploration capability. Second, an adaptive t-distributed feeding strategy was employed to define the connection between feeding behavior and temperature, increasing population variety and enhanced the algorithm's local search effectiveness. Finally, an adaptive ternary optimization mechanism is introduced in the exploration phase: a curve growth acceleration factor is used to collaboratively guide global and individual optimal information, while a hybrid adaptive cosine exponential weigh dynamically adjusts the search intensity. Additionally, an inverse worst individual variant reinforcement approach is employed to enhance the population evolution efficiency. In the hybrid test sets of CEC2005 and CEC2019, MSCOA shows improved convergence accuracy compared to the traditional COA algorithm, and the Wilcoxon test (p < 0.05) confirms its superiority over five other comparison algorithms. MSCOA outperforms other algorithms in terms of robustness, convergence speed, and solution accuracy, although there is still room for further improvement. When combined with Extreme Learning Machine (ELM) and applied to the Wisconsin breast cancer dataset, the MSCOA-ELM model achieved 100% accuracy and F1 score, a 28.9% improvement over the baseline ELM, demonstrating the algorithm's efficiency and generalization ability in solving practical optimization problems.