A filter-and-refine approach to lightweight application traffic classification

Abstract

As application traffic becomes increasingly complex and voluminous, the need for accurate and fast traffic classification is emphasized, leading to proposals for lightweighting DL-based classifier. Nevertheless, there is still a need for faster and more accurate classification methods for practical deployment. We propose a new traffic classification mechanism using the Filter-and-Refine approach. The proposed method was evaluated public dataset using seven baselines and showed 4%p higher accuracy and about 39 times faster classification speed compared to the state-of-the-art. The source code and dataset are available at https://github.com/pb1069/Network-Traffic-Classification.