Abstract

The Grey Wolf Optimizer (GWO) is a very recently developed and emerging swarm-intelligent algorithm. The GWO algorithm was inspired by the social dominance hierarchy and hunting strategy of the grey wolves that has been successfully tailored to tackle various discrete and continuous optimization problems. During its practical implementation, however, it may be stuck in sub-optimal solutions (stagnation in local optima) due to its less exploration in the early stages that show the main drawback of this algorithm. Therefore, this research work enhances the hunting and attacking mechanism in order to modify the corresponding position updated equation and exploitation equation, respectively, to propose a novel algorithm, called Weighted Grey Wolf Optimizer with Improved Convergence Rate (WGWOIC). The effectiveness of the proposed algorithm (WGWOIC) is investigated by testing it to 33 different and fairly popular numerical benchmark functions. Although, these test functions are considered from two different benchmark datasets to assess the strength and robustness of the proposed algorithm regarding the unknown search space of the problem. In order to the performance analysis, moreover, the WGWOIC’s results are compared against many other state-of-the-art meta-heuristic algorithms such as Particle Swarm Optimization (PSO), Moth-Flame Optimization (MFO), Whale Optimization Algorithm (WOA), Grey Wolf Optimizer (GWO), and very recent variants of GWO. The comparative study for WGWOIC concludes that the proposed algorithm provides very competitive results against other studied meta-heuristic algorithms. Furthermore, the hybridization of the WGWOIC meta-heuristic optimization algorithm with a Multi-Layer Perceptron (MLP) neural network is employed to improve the accuracy of the classification problem. WGWOIC trainer provides the optimal values for weight and biases to the MLP network. Further, the performance is tested in terms of classification accuracy on five popular classification datasets and assesses the efficiency of the WGWOIC trainer against many other meta-heuristics trainers. The results show that the proposed algorithm eventually provides very competitive outcomes, implying that the WGWOIC algorithm offers a better exploit and explores the search space and effectively solves several different classification problems.

Key words: Meta-heuristic Algorithms, Evolutionary Algorithms, Nature-Inspired Algorithms, Swarm-Based Algorithms, Grey Wolf Optimizer, Neural Network, Multi-Layer Perceptron.