A hybrid approach for mobile traffic classification: Integrating multimodal feature representation and machine learning

With the evolving developments in 5G networks and protocol technologies, and the complexity of mobile traffic classification demands an intelligent approach to handle the adaptive nature of 5G networks. The challenge even persists when integrating features from different modalities coexisting in flow-level and packet-level metadata. A unified feature representation can be generated using feature representation techniques of hierarchical learning models. By using the advantages of these representations, they can effectively capture packet, flow, and temporal features to obtain a rich fused latent representation of mobile traffic data. This unique approach can be utilized to train shallow and deep computational models to classify complex traffic patterns with high precision and accuracy. The hybrid approach is tested and evaluated on encrypted traffic patterns obtained from TLS and DTLS protocol with classification accuracy over 80% towards mixed traffic. The results obtained can improve network management and quality of experience in mobile networks.