Pruning algorithm in wavelet neural network for ECG signal classification

Wavelet neural networks have been widely studied in recent years, because they combine the adaptability of neural networks with the strong feature extracting ability of wavelet transforms. Because of the inevitable oscillatory behavior in wavelet functions, wavelet neural networks are susceptible to trap into local minima when using gradient descent training algorithms. In this paper, a pruning algorithm is introduced into wavelet neural networks for combating the problem of the gradient-descent algorithm, and its merits are analyzed. Good performance is obtained in experiments on ECG signal classification using the pruning algorithm.