Three-layer Feedforward Evolutionary Neural Network Method for Classification Based on Beetle Antennae Search Algorithm

Abstract

Artificial neural network depends largely on its parameters and structures, such as weights and biases in parameters. Most of the existing evolutionary neural network methods rely on different heuristic algorithms to optimize the neural network’s structure and parameters, such as particle swarm optimization, genetic algorithm, etc. However, due to the large population size of this method, a lot of computing resources are consumed for optimization, resulting in the low efficiency of the algorithm. In order to construct more efficient and accurate neural network, a beetle antennae search (BAS) based evolutionary neural network is proposed, which is called BASNN. In this algorithm, we firstly design a three-layer feedforward neural network with link switches for classification. Then, the structure of the neural network and its parameters are evolved simultaneously by beetle antennae search algorithm. We evaluate our algorithm on two practical classification problems and compare it with some related works. The result of this experiment shows that BASNN algorithm can construct a high precious neural network at very fast speed.