The Impact of Linear Transformation on SVM Margin

It is well recognized that support vector machines (SVM) would produce better classification performance in terms of generalization power. Based on the statistical learning theory (SLT), the margin scale reflects the generalization capability to a great extent. The bigger the margin scale takes, the better the generalization capability of SVM will have. This paper makes an attempt to investigate the impact of linear transformation on SVM margin. The linear transformation maps a linear vector space into another with dimension change. From the result of experiments, we know that the margin of SVM can be enlarged through the appropriate linear transformation. By specifying a particular linear transformation that is feature weight adjustment, the relative amount of margin between two data sets can be improved significantly.