Classification of encrypted networks in terms of content

The widespread use of the Internet today requires efficient and secure management of the existing computer network infrastructure. In addition, with the increase in the use of internet applications day by day, network traffic that creates large data volume is also emerging. In order to process big data, performance-oriented methods should be used. Network traffic data needs classification for applications in many workspaces, such as network management and security. Encryption of network traffic and applications such as VPN use complicate the network traffic classification process. In this thesis, a new performance-oriented platform has been developed for the classification of encrypted network traffic, which can be easily and quickly applied to real-time systems. Machine learning techniques were used in the classification process. Process management was carried out in order to apply experiment-based machine learning techniques effectively. Apache Spark for data processing, NFStream for feature extraction, and MLflow software technologies for process management were used in the design of the platform. In addition, this study has brought a new feature called "pattern byte" to the literature. Within the scope of the experiment carried out with the proposed platform, network traffic is classified by machine learning algorithms according to the application and application types. Performance results were evaluated as a result of using GBTree, LightGBM, and XGBoost algorithms from machine learning algorithms. Evaluation of performance results is examined by accuracy, recall, precision, and F1 scores. In the results examined, GBTree, LightGBM, and XGBoost algorithms achieve F1 scores of approximately 98%, 89%, and 99% in application classification. In classification according to application types, all algorithms reach 99% F1 scores. As a result, among the algorithms, it was seen that the XGBoost algorithm achieved the best result with an F1 score of over 99% in both classification problems.