Design And Applications Of Improved Metaheuristic Algorithms For Neural Network Training

Abstract

The success behind nature-inspired evolutionary metaheuristic algorithms lies in its seemly combination of operator s castoff for smooth balance between exploration and exploitation. The deficit in such combination leads to untimely convergence of an algorithm, simultaneously failed to attain global optimum by stocking in a local optimum. Salp Swarm Algorithm (SSA) and Spotted Hyena Optimizer (SHO) are two recently evolved optimization techniques, intended to resolve continuous, non-linear, and multifaceted real-world optimization complications. For solving complex day-to-day life problems the explorative strength of existing SSA and SHO is not adequate. This study introduces five improved hybrid versions of SSA and SHO, which improve the performance of the existing algorithm by using Space Transformation Search (STS), Oppositional Based Learning (OBL), Mutation Operator, and Quadratic Approximation Operator (QAO). The proposed algorithms are termed as STS-SSA, OBL-MO-SSA, STS-SHO, OBL-MO-SHO, and QASHO. The assimilation of STS, a new evolutionary technique enhances the exploration and exploitation capability in the search space and successfully avoids local optima entrapment. The oppositional learning concept ensures the current as well as opposite candidate solutions in the search region simultaneously to evaluate the closer solutions during the ongoing evolutionary process. The mutation operator avoids the arbitrary positions in the search region by choosing lesser and larger mutations for balanced motion in current and opposite directions. The proposed method OBL-MO-SSA and OBL-MO-SHO improve the exploration and exploitation strength inside the search region at the same time exhibiting better convergence speed by successfully avoiding local optima stagnation. The proposed QASHO has been scrutinized to enhance the exploitation ability, aiming to achieve global optimum, as QAO performs better within the local confinement region.