A Network Traffic Classification Method for Class-Imbalanced Data

Abstract

It is very common that flow distribution of class is not uniform in attack traffic. It wi11 lead to a 1ow classification accuracy in network intrusion detection. For overcoming this class imbalance phenomenon，a pipelining ensemble approach in different feature spaces was proposed，which translates multi-class classification to two-class classification. Based on the pipelining ensemble，it could be further conduct oversampling and customized feature selection for minority class，which may avoid the disturbance from majority class. The experiment result shows that the proposed approach can efficiently improve the accuracy of minority class of attack traffic.

Keywords

• attack traffic
• class imbalance
• classifiers pipelining ensemble