FEAML: A Mobile Traffic Classification System with Feature Expansion and Autonomous Machine Learning

Network traffic classification is a crucial component in network protocol and application identification, playing a pivotal role in various network and security-related activities. However, conventional traffic classification techniques are not suitable for mobile app traffic. This is primarily because mobile traffic exhibits a considerable disparity from Internet traffic, primarily in terms of the inconsistent traffic characteristics generated by the same app and imbalanced traffic samples, among others. To overcome these challenges, this paper presents a new mobile traffic classification system called FEAML(Feature Expansion and Autonomous Machine Learning), which leverages feature expansion and autonomous machine learning. The proposed system employs the SMOTE tool to address the imbalance problem and a hybrid architecture based on the self-attentive mechanism's CNN and Stacked LSTM layers to achieve feature expansion, thereby improving the generalization and classification accuracy of models. Through a series of data-driven experiments conducted on three public mobile traffic datasets, the proposed system demonstrates a 13% improvement in accuracy compared to state-of-the-art classification solutions.