Time-varying elite sand cat optimization algorithms for engineering design and feature selection

• Published in: Expert systems with applications Vol. 275; p. 127026
• Main Author: Zhang, Li
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 25.05.2025
• Subjects: Exploration and exploitation; Feature selection; Population diversity; Sand Cat Optimization Algorithm; Time-varying elites; Exploration and exploitation; Sand Cat Optimization Algorithm; Time-varying elites; Feature selection; Population diversity
• ISSN: 0957-4174
• DOI: 10.1016/j.eswa.2025.127026

The Sand Cat Optimization Algorithm (SCSO) is well known for its efficient search capability. However, SCSO has limitations in solving complex, high-dimensional problems, particularly in balancing exploration and exploitation and maintaining population diversity. It is prone to falling into local optimal solutions. This paper addresses these limitations by improving SCSO’s search strategies, enhancing its ability to solve complex high-dimensional problems and avoiding local optima. Here, we propose a Time-Varying Elite Sand Cat Optimization Algorithm (TVESCSO). This algorithm achieves a multi-strategy search by dividing the sand cat population into a new and traditional population and further dynamically dividing the traditional population into an optimal and free population. The optimal population focuses on global exploration and local exploitation guided by improved adaptive factors and elite individuals. In contrast, the free population conducts extensive searches guided by the elite pool, expanding the search scope and balancing exploration and exploitation dynamically. The new population’s diversity is further improved by merging the two sub-populations, information exchange and new population ordering. Finally, the elite individuals in the new population are mutated using the Logistic Chaos method, accelerating the algorithm’s convergence and helping to jump out of local optimal solutions. Experimental results show that TVESCSO exhibits faster convergence and higher accuracy in the benchmark functions of the Congress on Evolutionary Computation (CEC2017 and CEC2022), outperforming a wide range of state-of-the-art and well-known optimization algorithms. The Friedman Test , the Nemenyi and Wilcoxon Rank-Sum tests further validates the effectiveness and robustness of TVESCSO. The application of TVESCSO to four real-world engineering problems and twelve feature selection datasets further confirms its excellent performance. TVESCSO significantly improves SCSO performance and demonstrates significant advantages in complex numerical optimization, engineering optimization, and high-dimensional feature selection problems. This improvement addresses the issue of SCSO being prone to local optima and provides new insights for SCSO algorithm research, advancing the field further.