Decision-Improved Support Vector Machine and its Application

The practical applications of 1-SVM in pattern classification are limited due to the deficiency of its low classification precision. Aimed at solving this problem, first, a 1-DISVM is proposed, in which a coefficient is introduced to adjust decision curve region. Compared to 1-SVM, 1-DISVM inherits the ability to find outliers but gains improved classification accuracy by the coefficient adjusting. Based on 1-DISVM, an unsupervised learning multi-class classification model is also built. By means of small quantity of fault samples, the model improves classification performance by getting rid of the influence from wrong samples. Then the experiment is implemented by applying the considered approaches in gear-box fault diagnosis. Experimental results show that, the presented method achieved precise classification for two-class (normal and fault) data identification.