A Study on Classification of Mobile Application Network Traffic on MIRAGE-COVID-CCMA-2022 Dataset via Machine/Deep Learning and Feature Selection Methods

In recent years, the use of mobile internet has become widespread rapidly with the introduction of smartphones. The increasing weight of mobile network traffic in the overall network traffic has made mobile network traffic classification a new and important topic of study. The study utilizes MIRAGE-COVID-CCMA-2022 dataset for the classification of mobile application network traffic. The MIRAGE-COVID-CCMA-2022 dataset contains the network traffic of nine communication and collaboration mobile applications including Discord, GotoMeeting, Meet, Messenger, Skype, Slack, Teams, Webex, and Zoom. Experiments were conducted using four machine learning classifiers (Support Vector Machine, k-Nearest Neighbor, Adaptive Boosting, and Random Forest), four deep learning classifiers (Convolutional Neural Networks, Long Short-Term Memory, Bidirectional Long Short-Term Memory, and Gated Recurrent Units), and three feature selection methods (Minimum Redundancy Maximum Relevance, k-Nearest Neighbor, and Random Forest). The results of the study show that an overall accuracy of over 95% for the MIRAGE-COVID-CCMA-2022 dataset with a nine-class classification. The study concludes that the most appropriate classification structure is to select the features using the Minimum Redundancy Maximum Relevance method and classify the features with the Random Forest classifier. Also, this classification structure stands out as one of the classification structures with the fastest running time.