Adapting TSVM for fault diagnosis with imbalanced class data

Abstract

Support vector machines (SVM) have been used successfully in many classification problems. But many classification tasks involve imbalanced training examples in practice. Two novel algorithms are proposed to solve some imbalanced data classification by adapting transductive support vector machine (TSVM) and edited nearest neighbor (ENN) rules. Algorithm 1 chooses some useful test samples of positive class and adds them to the training sets. Those samples are used to add the lack of training samples. However, they may contain noisy examples. Therefore, edited nearest neighbor rule is removed the noisy examples. Algorithm 2 selects some useful testing documents of both positive class samples and negative class samples and then adds them to the training sets. Removing the noisy examples is similar to the first algorithm. Both of the two algorithms add the testing samples to maintain the balance between the minority class and the majority class. Simulation results show the feasibility and effectiveness of the proposed algorithm.