Support Vector Machine in Classification of
暂无分享,去创建一个
Positron annihilation spectroscopy is used in the study of radiation induced defects in nuclear materials in a non-intrusive way. The positron can be trapped by defects and the number of exponential components in the positron lifetime spectrum is related to the number of defect states. This work concerns classification of positron spectra with respect to the number of spectral components by support vector machine (SVM). The SVM has not yet been investigated for positron spectra analysis. The SVMs use an optimized generalization. The SVM classifier has been constructed by training with the simulated positron spectral data with two and three spectral components. Tuning the hyperparameters, such as the generalization parameter C, has been done by using the 10fold cross-validation error. The experimental spectra available from polymer materials have been analysed by the constructed SVM nonlinear classifier. Experimental data are classified as the class of positron spectra with three spectral components with accuracy of 95.4 %. The SVM calculations show that certain degree of misclassification tolerance can produce a solution with good expected generalisation.
[1] R. Chakarova,et al. Unfolding positron lifetime spectra with neural networks , 1999 .
[2] Imre Pázsit,et al. Analysis of the experimental positron lifetime spectra by neural networks , 2003 .
[3] Vladimir N. Vapnik,et al. The Nature of Statistical Learning Theory , 2000, Statistics for Engineering and Information Science.
[4] Vladimir Cherkassky,et al. Learning from data , 1998 .
[5] I. Procházka,et al. Positron Lifetime Study of Reactor Pressure Vessel Steels , 2000 .