Quantitative analysis of volatile organic compounds using ion mobility spectrometry and cascade correlation neural networks

Ion mobility spectrometry (IMS) has a limited linear range. Nonlinear calibration methods, such as neural networks are ideally suited for IMS due to their capability of modeling complex systems. Many neural networks suffer from long training times and overfitting. Cascade correlation neural networks (CCN) are interesting, because they train at fast rates. Another advantage of CCNs is that they automatically configure their own topology (number of layers and number of units in each layer). By using a the decay parameter in training neural networks, reproducible and general models may be obtained at the cost of longer training times. CCN networks were trained to furnish both quantitative and qualitative prediction for a complex IMS data set (229 spectra, 200 input points, and 15 output classes). The advantage of rapid training is that replicate neural networks may be obtained. The precision of replicated network predictions appears to provide a measure of accuracy. Partial least-squares regression (PLS) is used as a comparative method. The CCN with decay rates an order of magnitude larger than learning rate achieves significantly better results than those obtained from an optimal PLS model.