Real-Time VoIP Classification in Encrypted Traffic Using Deep Learning

With over 95% of internet traffic now encrypted, traditional methods like Deep Packet Inspection (DPI) and payload-based classification have been rendered ineffective. This study investigates the application of deep learning for real-time Voice over IP (VoIP) classification in encrypted network traffic. The ineffectiveness of traditional methods motivates the exploration of deep learning models, including Long Short-Term Memory (LSTM), Multi-Layer Perceptron (MLP), and 1D Convolutional Neural Network (1D-CNN). This work leverages the pattern recognition power of deep learning to classify internet traffic as either VoIP or non-VOIP. The study includes improving data diversification by collecting and integrating network traces of VoIP and non-VoIP traffic from different origins. Furthermore, the efficacy of manual and automatic feature extraction techniques is comparatively evaluated. The findings demonstrate that manually extracted features yield superior classification accuracy. Manual feature extraction resulted in LSTM, MLP, and ID-CNN accuracy rates of 94.7%, 94.8%, and 67.5%, respectively. Automatic feature extraction yielded 73.1%, 71.6%, and 41.5%. Notably, the MLP model achieves the highest accuracy (94.8%) while maintaining significantly faster classification times. This efficiency makes it well-suited for real-time VoIP classification.