A Dual-Momentum Hybrid Wavelet Neural Net (DM-HWNN): Its performance evaluation and application

Abstract

This paper presents a novel single neural net-based classifier called Dual-Momentum Hybrid Wavelet Neural Nets (DM-HWNN). DM-HWNN inherits capability in learning efficiency from Wavelet Neural Networks (WNN) and performance consistency in classification from Back-Propagation Networks (BPN). An extra momentum term is introduced into the learning process to further speed up the convergence of the learning. K-fold cross validation (CV) over four benchmark datasets are conducted to compare the performance of this single neural net classifier with some 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 DM-HWNN outperforms the first three methods in term of classification accuracy and the SONG in term of computation time. Furthermore, a cutter dataset from industry milling machine is used to evidence classification capability of DM-HWNN and illustrate how DM-HWNN can be used in prediction of cutter's wear out.