Abstract
The efficiency of the swarm-based approach depends on the perfect balance of operators: exploration and exploitation. Due to a lack of balance between these two factors, the Salp Swarm Algorithm (SSA), a recently developed swarm-based metaheuristic approach, suffers from confined minima, stuck and untimely premature convergence. This paper introduces a new, improvised hybrid SSA named Local Search SSA (LS-SSA) to address the pitfalls associated with standard SSA. To prove the competency of the suggested LS-SSA, it is assessed over the twenty-eight functions suite of IEEE-CEC-2017 relating to a diverse set of contemporary methods. Furthermore, a sequence of non-parametric assessments was conducted to establish the statistical significance of the proposed LS-SSA. As a weak exploitation strength for neighbor exploration, SSA may result in less refined parameter tuning of CNN for healthcare-based medical imaging. Hence, LS-SSA is an effective algorithm for hyperparameter tuning of CNNs evaluated on medical imaging datasets, specifically brain MRI. This leads to improved model performance, characterized by higher accuracy, reduced standard deviation, lower minimum RMSE values, and higher average performance. Consequently, optimal candidate solutions with improved and faster convergence toward global optima are achieved.