An Efficient Approach Based on Parameter Optimization for Network Traffic Classification Using Machine Learning

Different kinds of attacks on the network have greatly increased due to the exponential increase in the number of users over the last decade. This has greatly hampered transactions online especially when they are financially based. Hence, there is every need to develop and design new cybersecurity techniques to curb these cyber-crimes. This study presents the accurate classification of the Transmission Control Protocol (TCP) and User Datagram Protocol (UDP) flows on the network using Machine Learning (ML) based on parameter optimization. The method was validated with three randomly selected Machine Learning (ML) algorithms of the Waikato Environment for Knowledge Analysis (WEKA) software with 10 folds cross-validation and chose the algorithm with the best performance. Experiments were carried out on three scenarios using the USTC-TFC2016 dataset and the results demonstrated that Random Forest (RF) achieved average performance accuracy of 99.52%, which is 0.11% higher than the the-state-of-the art approach. It was also shown that the Decision Tree (J48) achieved higher classification accuracy than Naive Bayes (NB).