Theoretical aspects of mapping to multidimensional optimal regions as a multi-classifier
Intelligent Data Analysis
Hi-index | 0.00 |
In this paper, we present a new method of using Support Vector Machine (SVM) for multiclass classification. In our method, we use a tree based SVM classifier for classification. Compared with the other SVM multi-class classification methods in literature (i.e. One-Against-One, DAGSVM), our proposed SVM tree classifier is more efficient in both training/classification. Our new SVM tree classifier requires o(n) SVM training during the training stage and o(log(n)) SVM testing during the test stage, while other methods require o(n^2) or at best o(n) SVM training during the training and o(n^2) or at best o(n) SVM testing during testing. Experimental results on digital library document classification demonstrate that our methods is not only significantly more efficient but also achieves the similar precision of classification.