Parameter optimization of evolving spiking neural networks using improved firefly algorithm for classification tasks

Abstract

Evolving Spiking Neural Network (ESNN) is the third generation of artificial neural network that has been widely used in numerous studies in recent years. However, there are issues of ESSN that need to be improved; one of which is its parameters namely the modulation factor (Mod), similarity factor (Sim) and threshold factor (C) that have to be manually tuned for optimal values that are suitable for any particular problem. The objective of the proposed work is to automatically determine the optimum values of the ESNN parameters for various datasets by integrating the Firefly Algorithm (FA) optimizer into the ESNN training phase and adaptively searching for the best parameter values. In this study, FA has been modified and improved, and was applied to improve the accuracy of ESNN structure and rates of classification accuracy. Five benchmark datasets from University of California, Irvine (UCI) Machine Learning Repository, have been used to measure the effectiveness of the integration model. Performance analysis of the proposed work was conducted by calculating classification accuracy, and compared with other parameter optimisation methods. The results from the experimentation have proven that the proposed algorithms have attained the optimal parameters values for ESNN.