Wavelet transform preprocessing for temperature constrained cascade correlation neural networks

Wavelet transform (WT) preprocessing offers two advantages: data compression and noise reduction. Wavelet compression increases the training rate of a neural network and allows neural network models to be obtained from data that otherwise would be prohibitively large. Two types of WT compressions have been studied: linear and nonlinear. Linear wavelet compression in which data are compressed by frequency usually provides better compression efficiency. Nonlinear wavelet compression in which data are compressed by amplitude is useful when the information cannot be easily represented by low-frequency components. The reduction of noise is important in the multivariate analysis because many methods overfit the data if care is not taken. Ion mobility spectrometry (IMS) is a sensing technique that can generate large amounts of data in short-time monitoring events. The temperature constrained cascade correlation neural networks (TCCCN) are a powerful chemometric method for pattern recognition. The IMS data of some volatile organic compounds are used to evaluate the WT-TCCCN method, and the results indicate that WT-TCCCN works well.