Improving SVM through a Risk Decision Rule Running on MATLAB

摘要

Support Vector Machine (SVM) is a classificationtechnique based on Structural Risk Minimization (SRM),which can run on MATLAB. For classification ofnonseparable samples, conventional SVM needs to select atradeoff between maximization the margin andmisclassification rate. In order to guarantee generalizedperformance and low misclassification rate of SVM, thispaper puts forward an improved SVM through a riskdecision rule for the nonseparable samples running onMATLAB. The improved SVM transforms the outputs ofthe SVM to posterior probabilities belonging to differentclasses and samples between the support hyper-planes areclassified by using risk decision rule of Empirical RiskMinimization (ERM). Computational results show that theproposed approach is better than conventional SVMremarkably when the two classes are easy to separate, andin other condition, its performance is comparable toconventional SVM.