Quantized Orthogonal Experimentation SSA (QOX-SSA): A Hybrid Technique for Feature Selection (FS) and Neural Network Training

The standard metaheuristics are commonly based on a single metaphorical model miming a particular animal group's food-searching behaviour. As a result, the contributions of such approaches are becoming inadequate in dealing with the evolving complexities of optimizing objective functions. Salp Swarm Algorithm (SSA) is one of them, a new swarm intelligence-based technique relying on a lone metaphor introduced for tackling global optimization issues. Nonetheless, SSA has garnered substantial acknowledgement and attraction among the research community because it is easy to implement and requires few control parameters to fine-tune. However, the typical SSA encounters confinement issues in local optima and an insufficient convergence pace when confronted with more intricate scenarios due to deficient population diversity, local exploitation, and global exploration. Therefore, this research integrates a Quantized Orthogonal Experimentation (QOX) operator to enhance population variety and intensify SSA's local exploitation and global explorative potential. The resulting hybrid approach is named QOX-SSA. QOX-SSA's optimization skill is demonstrated using 14 fundamental and 30 advanced benchmark problems of IEEE-CEC-2014 and comparing its effectiveness to some contemporary metaheuristics. Three nonparametric tests are conducted to verify the statistical importance of QOX-SSA's results. Furthermore, the applicability of QOX-SSA is examined by implementing it to train the Radial Basis Function Neural Network to classify data and resolve problems related to optimal feature selection. Experimental outcomes of QOX-SSA over different optimization issues confirm its superior performance compared to SSA and the alternate metaheuristics used for comparative analysis.