Abstract
The basic Salp Swarm Algorithm (SSA) offers advantages such as a simple structure and few parameters. However, it is prone to falling into local optima and remains inadequate for seed classification tasks that involve hyperparameter optimization of machine learning classifiers such as Support Vector Machines (SVMs). To overcome these limitations, an Enhanced Knowledge-based Salp Swarm Algorithm (EKSSA) is proposed. The EKSSA incorporates three key strategies: Adaptive adjustment mechanisms for parameters c1 and α to better balance exploration and exploitation within the salp population; a Gaussian walk-based position update strategy after the initial update phase, enhancing the global search ability of individuals; and a dynamic mirror learning strategy that expands the search domain through solution mirroring, thereby strengthening local search capability. The proposed algorithm was evaluated on thirty-two CEC benchmark functions, where it demonstrated superior performance compared to eight state-of-the-art algorithms, including Randomized Particle Swarm Optimizer (RPSO), Grey Wolf Optimizer (GWO), Archimedes Optimization Algorithm (AOA), Hybrid Particle Swarm Butterfly Algorithm (HPSBA), Aquila Optimizer (AO), Honey Badger Algorithm (HBA), Salp Swarm Algorithm (SSA), and Sine-Cosine Quantum Salp Swarm Algorithm (SCQSSA). Furthermore, an EKSSA-SVM hybrid classifier was developed for seed classification, achieving higher classification accuracy.