Graph neural network based network traffic classification

Abstract

a brief introduction of GNN and what actually GNN is. Then we focused on general design pipeline of GNN by describing all the four steps along with a pictorial representation of the same. Then we give some information about the types of GNN following the graph neural network model. After that, we studied graph embeddings, it’s advantages as well as applications. Then we go for working of GNN i.e how actually GNN works. When one try to analyze a graph; it is not that much easy and we try to focus on the challenges to be faced. In the later section, we have discussed about Graph Convolutional Network along with it’s type and how it is different from GNN. The learning/training process of neural network is broadly divided into three types i.e Supervised, Memory-based and Unsupervised learning. Two machine learning algorithm- KNN algorithm and Back-Propagation algorithm is descried. Lastly we have discussed about some applications of GNN from theoretical point of view and after that we also studied some real life applications of GNN in today’s world. Next, we give our attention to network traffic classification which is no doubt a very important topic in today’s scenario where cyber crime is increasing rapidly day by day. In the early days of internet, some traditional traffic classification method is used. But, all of them have some drawbacks and to overcome those researchers have introduced machine-learning in traffic classification. However, in this work we are mainly focused on how machine learning can help to classify a particular network traffic flow. In this thesis work, we have tried to build a GNN model that can be used to classify network traffic. This technique relies on the extraction of topological data in the network traffic. Experiences from the experiments have shown GNN based network traffic classifier as a promising and effective solution. Further a thorough assessment of the proposed technique using actual datasets would validate its acceptability compared to state-of-the-art techniques.