A direct method of nuclear pulse shape discrimination based on principal component analysis and support vector machine

Fault diagnosis and particle discrimination can be fundamentally solved as a case of pulse shape discrimination (PSD). The classical methods of PSD are inconvenient or not effective when more than two pulse shapes need to be discriminated or the pulse shapes have only small differences. A direct method to discriminate nuclear pulse shapes based on principal component analysis (PCA) and support vector machine (SVM) is reported in this paper. The training and testing accuracies of SVM classifiers with different kernels were not the same, and the algorithms were shown to have great noise immunity. Though the samples in the Group A and Group C cannot be discriminated with the naked eye, the accuracies are all above 94.7% if suitable SVM kernels are selected. There is no evidence showing that the Gaussian kernel is superior. The lower sampling frequency of the analog-to-digital converter and the information loss caused by dimension reduction were also considered.