Towards Accurate Categorization of Network IP Traffic Using Deep Packet Inspection and Machine Learning

Network traffic classification is crucial for optimal network resource management. Several network traffic classification methods have been proposed, e.g., Deep Packet Inspection (DPI), and machine learning-based network traffic classification. Each approach is generally efficient for a certain class of network traffic. However, there is no one-fit-all method, i.e., no method offers the best performance for all types of network traffic. In this paper, we propose a hybrid network traffic classification technique that uses a combination of DPI and machine learning to identify and classify the network traffic into different Quality of Service (QoS) classes. The traffic is first identified through the DPI module, and the unidentified traffic then goes through the machine learning module, offering a classification accuracy of more than 98%. The results are evaluated based on the combination of DPI and different machine learning methods, e.g. supervised and unsupervised learning algorithms.