Improving the discriminatory capabilities of a neural classifier by using a biased-wavelet layer

In the context of wavelet neural networks (WNN's), two modifications to the basic training algorithms are proposed, namely the introduction of a bias component in the wavelets and the adoption of a weight decay policy. A problem of ECG segment classification is used for illustration purposes. Results suggest that bias improves the discriminatory capabilities of the WNN, which is also compared favourably to a conventional perceptron classifier. The use of weight decay during training, followed by pruning, resulted in a more parsimonious network, which also turned out to be a more conservative classifier. The knowledge embedded in the wavelet layer is interpreted with basis on the concept of super-wavelets.