Generalization Analysis of Neural Networks for Gas Impurity in Air

The support vector machine was adopted to recognize the nonlinear fluorescence spectrum after compressed by wavelets transform. In order to investigate the generalization capability of neural network more roundly, a model for the testing data is proposed. The generalization capability of the support vector machine (SVM) network of this work and that of the probabilistic neural network (PNN) of a previous work are compared with the data produced by the model. The simulation results show that the SVM network provides better generalization capability than that of the PNN network for either laboratory data or changes data in experimental conditions.