A Fuzzy Support Vector Machine for Imbalanced Data Classification

The usual fuzzy support vector machines are often affected by the number and distribution of data samples. In order to solve the existing problems, a fuzzy membership is proposed and then a new fuzzy support vector machine was constructed, which is suitable for imbalanced number and distribution data sets. The results show that for welding defects data set welding1, the proposed algorithm under different parameters is superior to the traditional algorithms of SVM and FSVM, whose classification error rate and bias are lower and less affected by parameters; for usual data sets sonar, diabetes, parkinsons, the proposed algorithm has better performances on the classification balance and stability, and its training time is acceptable, which shows this algorithm has good versatility.