ARTIFICIAL FISH SWARM OPTMIZATION FOR MULTILAYERNETWORK LEARNING IN CLASSIFICATION PROBLEMS

Abstract

Nature-Inspired Computing (NIC) has always been a promising tool to enhance neural network learning. Artificial Fish Swarm Algorithm (AFSA) as one of the NIC methods is widely used for optimizing the global searching of ANN. In this study, we applied the AFSA method to improve the Multilayer Perceptron (MLP) learning for promising accuracy in various classification problems. The parameters of AFSA: AFSA prey, AFSA swarm and AFSA follow are implemented on the MLP network for improving the accuracy of various classification datasets from UCI machine learning. The results are compared to other NIC methods, i.e., Particle Swarm Optimization (PSO) and Differential Evolution (DE), in which AFSA gives better accuracy with feasible performance for all datasets.