Multimedia Traffic Classification for Imbalanced Environment

With ever-increasing volume and variety of multimedia traffic on the Internet, machine learning-empowered techniques nowadays tend to become indispensable for future intelligent network management. To realize automatic traffic management with Quality of Service (QoS) guarantees, there is a pressing need for accurate traffic classification. However, the inherent characteristics of networks cause imbalanced class distribution in traffic classification, which could degrade the performance of classification, especially on the minority classes. To address the issue of class imbalance in both stationary and nonstationary environments, this paper proposes a novel scheme called CHS (chain hierarchical structure) which is able to characterize class distribution from a new perspective. By building an error model, we can compute the error propagation generated by CHS and analyze the factors that affect it. More importantly, two key methods involving classifier ranking and combination with the hierarchical structure are devised to mitigate the error propagation produced by the classifier. The effectiveness of the developed framework is validated through experiments over two real-world traffic datasets in both stationary and nonstationary environments. The experimental results demonstrate that our proposed methods outperform the state-of-the-art approaches in terms of classification accuracy and running time. The proposed methods are particularly effective in the nonstationary imbalanced environment.