Firefly based ridge polynomial neural network for classification

Abstract

Classification using higher order neural network (HONN) such as pi-sigma and ridge polynomial neural network (RPNN) are the most salient and active research area and popularly used in several applications such as financial time series forecasting and for solving inverse problems in electromagnetic non-destructive evaluation. This paper intends to use RPNN for classification which overcomes certain limitations of MLP having slow learning properties and ability to get stuck in local minima. RPNN distinguish themselves from MLP due to their fast learning capability and powerful mapping of single layer trainable weights in networks. Firefly algorithm (FFA) is used for training of the RPNN and then the proposed technique is tested with three different real world dataset such as, glass, iris and Haberman's survival datasets archived from UCI respiratory. The Simulation results shows that the classification accuracy and the convergence rate of FFA based RPNN is higher as compared with FFA based MLP