Selective SVM Ensemble Base on Clustering Analysis Apply for Analog Circuit Fault Diagnosis with Small Samples

Abstract

This paper presents a selective SVM ensemble based on clustering analysis to localize the faults of analogue circuits with small samples. The method overcomes disadvantages of single SVM and greatly improves the generation ability for problems with small samples. In the end of paper, simulation experiments on a CTSV filter circuit are carried out. Experimental results demonstrate that selective SVM ensemble based on clustering algorithm is a more effective method to fault diagnosis with small samples than single SVM.