Abstract:In fault diagnosis application, the performance of traditional support vector machine (SVM)drops significantly when it is applied to the problem of learning from imbalanced datasets where the fault instances heavily outnumbers the normal instances. To address this problem, a novel fault detection SVM approach is proposed which is based on spectral clustering combined with SVM under unbalanced samples. In order to classify the unbalanced samples correctly. Majority instances are clustered using spectrum cluster in kernel space for resampling reprentative samples, so as to balance the training samples and enhance the classification performance. The proposed algorithm is applied in fault detection of bearings and is compared against other methods. The experimental results show that our approach achieves better detection performance than other methods.