Improved whale trainer for sonar datasets classification using neural network

Abstract

To classify various sonar dataset, this paper proposes the use of the newly developed Whale Optimization Algorithm (WOA) algorithm for training Multi-Layer Perceptrons Neural Network (MLPs NN). Similar to other evolutionary classifiers, trapping in local minima, slow convergence rate, and non-real-time classification are three shortcomings it confronts in solving high-dimensional problems. Due to the novelty of WOA trainer, there is little in the literature regarding decreasing aforementioned deficiencies. In this paper, we also utilize seven spiral shapes to improve the performance of the WOA trainer. To assess the performance of the proposed classifiers, these networks will be evaluated using the three real-world practical sonar dataset. For endorsement, the results are compared to five popular meta-heuristics trainers include Particle Swarm Optimization (PSO), Gravitational Search Algorithm (GSA), Ant Colony Optimization (ACO), Gray Wolf Optimization (GWO), and WOA. The results show that new classifiers indicate better performance than the other benchmark algorithms, in terms of avoidance from getting stuck at local minima, classification accuracy, and convergence speed.