MACHINE LEARNING-BASED SELECTION OF INFORMATIVE FEATURES FOR NETWORK TRAFFIC CLASSIFICATION

Abstract

Internet traffic is rapidly increasing as mobile and web applications become more complex and often generate multiple service flows simultaneously, reducing the effectiveness of traditional port-based and DPI approaches. Machine learning enables traffic classification using features derived from packet headers and flow-level statistics; however, not all extracted features contribute equally. Some features improve accuracy, while others introduce noise and computational overhead. Therefore, selecting informative features is crucial for accurate and efficient classification, particularly in near real-time monitoring and traffic management systems. This paper reviews three feature selection families filter, wrapper, and embedded methods and summarizes their main ideas, strengths, and limitations.