A novel Hybrid Wavelet Neural Network and its performance evaluation

Abstract

In this paper, we propose a novel single neural net-based classifier called hybrid wavelet neural networks (HWNN). HWNN makes good use of the characteristics of wavelet neural networks (WNN) and back-propagation neural networks (BPN), so that it inherits WNNpsilas capability in learning efficiency and BPNpsilas performance consistency in classification problems. We conducted k-fold cross validation (CV) to compare the performance of this single neural net classifier with some of the other existing multiple classifier systems (MCS) including logiboost Bayesian classifier (LBC), multistage neural networks ensemble (MNNE), and self-organizing neural Grove (SONG). The results show that HWNN achieves higher classification accuracy than the three methods being compared. Besides, HWNN is faster than SONG in terms of computation time. Furthermore, we augment HWNN by introducing an extra moment term in the learning process to further speed up the convergence of the learning. Both experimental and theoretical results demonstrate that this improved HWNN (iHWNN) actually outperforms HWNN in terms of classification accuracy and computation time.