Adaptive Dimensional Search Based Orthogonal Experimentation SSA (ADOX-SSA) for training RBF Neural Network and optimal Feature Selection

Abstract Salp Swarm Algorithm (SSA) is a new stochastic approach for solving optimization issues based on the ideas of swarm intelligence. The ease of implementation and lower number of parameters to fine-tune are responsible for SSA's success and universal acceptance among researchers. The typical SSA method suffers from local optima entrapment and a poor convergence rate while dealing with more complex situations because of a lack of population density and inadequate local exploitation. To overcome such challenges, this study suggests a hybrid kind of SSA called Adaptive-Dimensional-Search based Orthogonal Experimentation SSA (ADOX-SSA). Furthermore, the inclusion of an ADOX operator increases population diversity, intensifies local exploitation, and strengthens the SSA standards. Consequently, the balance between the processes of exploration and exploitation is attuned, resulting in a faster rate of convergence than the normal SSA. To validate the ADOX-SSA technique's competency, 14 basic functions and 30 advanced standard functions were chosen following IEEE-CEC-2014. The findings of the proposed strategy have also been compared to those of recent metaheuristic approaches. Two nonparametric tests were employed to demonstrate statistical significance as Friedman and Holms approach. Additionally, the suggested ADOX-SSA approach is castoff for training Radial Basis Function Neural Network (RBFNN) by selecting datasets from the UCI depository. Finally, the same suggested approach is utilized to select the best features from benchmark datasets while maintaining accuracy and minimizing neural network complexity.