Network traffic classification based on improved DAG-SVM
Shengnan Hao, Jing Hu, Songyin Liu, Tiecheng Song, Jinghong Guo, Shidong Liu · 2015
Network traffic classification plays a fundamental role in network management and services. Given error accumulation in traditional DAGSVM (Directed Acyclic Graph-Support Vector Machine) algorithm, we propose an improved DAGSVM classification method using two different possibility metrics in this paper. Differing from traditional DAG-SVM, the improved DAG-SVM algorithm eliminates one class only under the condition of that classification error probability is less than threshold. The experiment results show that compared with traditional DAG-SVM, the methods proposed in this paper both have higher classification accuracy with acceptable time cost and improved DAG-SVM based on distance has a better performance than improved DAG-SVM based on decision function.