Optimizing Feature Selection for Efficient Encrypted Traffic Classification: A Systematic Approach

Abstract

Traffic classification is a technology for classifying and identifying sensitive information from cluttered traffic. With the increasing use of encryption and other evasion technologies, traditional content- based network traffic classification becomes impossible, and traffic classification is increasingly related to security and privacy. Many studies have been conducted to investigate traffic classification in various scenarios. A major challenge to existing schemes is extending traffic classification technology to a broader space. In other words, most traffic classification work is not universal and can only show great performance on specific datasets. In this article, we present a systematic approach to optimizing feature selection for encrypted traffic classification. We summarize the optional encrypted traffic features and analyze the approaches of feature selection in detail for different datasets. The experimental result demonstrates that our scheme is more accurate and universal than other state-of-the-art approaches. More precisely, our mechanism provides a guideline for future research in the field of traffic classification.