Neighborhood Based SVM Multi-classification Method for Condition Assessment of Insulator

Support vector machines (SVMs) are a class of popular classification algorithms for high generalization ability. However they mainly solve with two-classification problem, while, in practice, there are lots of multi-classification problem still. Based on advantages and shortcomings of existing multi-classification, a kind of neighborhood based SVM multi-classification method is proposed in this paper. To classify Samples in K classes just need to construct K-1 SVM classifiers with this method. And classifiers at prediction stage are chosen according to neighborhood of test samples. By using neighborhood based SVM and pair wise SVM to train these five stages and forecast respectively, the usefulness of this multi-classification method and more efficient is proved. In addition, an insulator condition valuation model based on neighborhood is obtained.