Construction of Binary Tree Classifier Using Linear SVM for Large-Scale Classification
Qiangkui Leng, Shurui Wang, Dehai Shen · 2018
Support vector machines (SVM) with kernel can solve nonlinear problem, but when the size of the problem is relatively large, the solving speed will be slow, which is not conducive to real-time applications. For linear SVM, it has fast computational speed, but its classification accuracy is usually not guaranteed. This paper proposes a binary tree classifier with linear SVM, which makes a tradeoff between computational speed and classification accuracy. If the local error rate is below a pre-set threshold, leaf nodes that make the final decision are generated; Otherwise, recursive construction of non-leaf nodes is performed. The final tree structure expresses the hierarchical division of given pattern classes. Experiments show that the proposed method ensures the generalization ability while responding rapidly.