Prediction Method of Infection Spreading with CNN for Self-evolving Botnets

Recently, machine learning has been extensively used and achieved significant results in many research areas. Accordingly, the literature has suggested the appearance of self-evolving botnets, which autonomously discover vulnerabilities and evolve by performing machine learning with computing resources of zombie computers. Our previous work have shown that the infection dynamics of self-evolving botnets depend on connection relations among hosts, through simulation experiments based on a Markov chain. This paper proposes a prediction method of the infection dynamics of self-evolving botnets, which uses a convolutional neural network (CNN). The proposed method predicts the level of infection spreading of self-evolving botnets, which depends on network structures and initial infected hosts, by using adjacency matrices of hosts as input data to CNN. In this paper, we show the effectiveness of the proposed method through performance evaluation based on data obtained from simulation experiments.