Abnormality Detection in Network Traffic by Classification and Graph Data Analysis

The invention of the Internet, along with the technology industry innovation, accommodates human society in the new age conveniently and effortlessly. Cyber-attacks, however, have increased drastically in the forms of ransomware, crypto- jacking, or insider threat. Many efforts have provided ways against network attacks as they put the digital world on the cliff edge. In this paper, we identify the abnormalities in the network traffic using a machine learning classification model and visualize the results via the graph network. We balanced the data, applied random forest to the training data, and used testing data to assess prediction accuracy. One novelty of our research is to utilize graph networks to visualize the ground truth and predicted network traffic. Both numerical performance and graph visualization support the competitive performance of our learning method.