An improved P-SVM method used to deal with imbalanced data sets

Abstract

Potential Support Vector Machine (P-SVM) is a novel Support Vector Machine (SVM) method. It defines a new optimization model which is different from standard SVM. However, P-SVM method has restrictions in dealing with unbalanced data sets. To solve this problem, an improved P-SVM method used to deal with imbalanced data sets is proposed in this paper. By using different penalty parameters to different slack variables in P-SVM, the new algorithm adjusts penalty parameters more flexible, and effectively improves the low classification accuracy caused by imbalanced samples. From theoretical analyses and experimental results, they have shown that this new method can obtain better classification accuracy than standard SVM and P-SVM in dealing with imbalanced data sets.