Evaluation of Randomized Variable Translation Wavelet Neural Networks

Abstract

A variable translation wavelet neural network (VT-WNN) is a type of wavelet neural network that is able to adapt to the changes in the input. Different learning algorithms have been proposed such as backpropagation and hybrid wavelet-particle swarm optimization. However, most of them are time costly. This paper proposed a new learning mechanism for VT-WNN using random weights. To validate the performance of randomized VT-WNN, several experiments using benchmark data form UCI machine learning datasets were conducted. The experimental results show that RVT-WNN can work on a broad range of applications from the small size up to the large size with comparable performance to other well-known classifiers.