Design and implementation of a neighborhood search biogeography-based optimization trainer for classifying sonar dataset using multi-layer perceptron neural network

Abstract

Multi-Layer Perceptron Neural Network (MLP NN) is one of the most applicable tools in solving complicated problems as well as classifying between target and non-target in sonar applications. In this paper, we use Biogeography-based Optimization (BBO) to train the MLP NN. Due to improving the exploration ability and enhancing the diversity of the population, we propose mutation operators into BBO and call it Neighborhood Search Trainer (NST). In addition, these operators prepare more balance between exploration and exploitation ability of BBO and induce it to record best results for solving high-dimensional problems. To assess the performance of the proposed classifier, this network will be evaluated with three datasets with various sizes and complexities. The results are compared with some of the most popular meta-heuristic algorithms for verification. The simulation results show that the new classifier performs better than the other benchmark algorithms and also than original BBO in terms of avoidance trapping in local optima, classification accuracy, and convergence speed. This paper also implements the designed classifier on the FPGA substrate for testing the real-time application of the proposed method.