Identifying Propagation Source of Worms with Convolutional Neural Networks
Can Yang Zhang, Peng Chao Zhou · 2022
Internet worms always make global impacts over the cyber space by network propagations. As a result, identifying the propagation source of the worms plays a very important role in the later worm analysis and forensics. In the paper, we propose a new source identification method using convolutional neural network (CNN). In particular, we consider the adjacent matrix of propagation graphs as the inputs to the CNN model, while output the source node as the model's class label. In this way, we can solve the source identification problem from the view of propagation graph classifications. We understand the CNN model may render to be ineffective when there exist a large number of source nodes to be classified, and thereby deploy our CNN model in a hierarchical architecture in order to reduce the number of class labels for CNN classification. To demonstrate the effectiveness of our model, we have conducted extensive experiments on both the synthetic and real-world networks. Our results have successfully confirmed the better identification performance of our CNN model than the competing solutions in the literature.