Encrypted Network Traffic Classification Using Intelligent Techniques

Traffic classification is considered one of the central components of network management and security operations, as it covers significant aspects such as traffic prioritization, anomalous behavior identification, and security policy implementation. Methods for traffic classification based on port and payload analysis are becoming less effective due to the nature of contemporary applications: dynamic ports, encrypted traffic, and complex protocols. To address these issues, two promising approaches have gained popularity: the use of machine learning and deep learning methods, which allow the identification of traffic sources based on patterns rather than relying on protocols and ports. In this study, the efficiency of various models is assessed, including support vector machines, random forests, convolutional neural networks (CNNs), recurrent neural networks (RNNs), CNN-RNN, and long short-term memory networks. This work indicates that deep learning models such as CNNs and RNNs have outperformed traditional machine learning models. Among them, the model based on the RNN architecture achieved the highest accuracy of 93% across all datasets. The random forest model also performed strongly, with notable improvements in precision and recall. This paper offers important information that will be useful for improving existing models and developing new models for network traffic classification, which is the basis of intelligent security systems.