The Effect of Convolutional Neural Network Layers on Payload-Based Traffic Classification

Different applications in modern networks produce various types of traffic with diverse service requirements. In the network traffic classification, "unknown applications" are regarded as a difficult problem that remains unsolved, especially in the healthcare sector. Traffic classification helps in classifying and aggregate traffic flows into categories with the same traffic patterns. Identification and classification of traffic are critical for network management efficiency, which includes Quality of Service (QoS), detection of intrusions, and lawful interception. Only the network traffic classification technology based on payloads is fitting because most of the applications are IP based, whether is attached to a specific port number or is dynamic or is temporary. Payload-based classifiers consist of finding the features in the payload of data packets to differentiate between the application protocols. In this work, we propose a model using machine learning (ML) for an accurate and efficient traffic classification. ML allows for an automatic response to various applications by classifying traffic without a network operator interference. Experimental results demonstrate that ML-based traffic classification methods are effective and obtained high accuracy and a low data loss rate in front of other available models.