Nexat: a history-based approach to predict attacker actions

Computer networks are constantly being targeted by different attacks. Since not all attacks are created equal, it is of paramount importance for network administrators to be aware of the status of the network infrastructure, the relevance of each attack with respect to the goals of the organization under attack, and also the most likely next steps of the attackers. In particular, the last capability, attack prediction, is of the most importance and value to the network administrators, as it enables them to provision the required actions to stop the attack and/or minimize its damage to the network's assets. Unfortunately, the existing approaches to attack prediction either provide limited useful information or are too complex to scale to the real-world scenarios. In this paper, we present a novel approach to the prediction of the actions of the attackers. Our approach uses machine learning techniques to learn the historical behavior of attackers and then, at the run time, leverages this knowledge in order to produce an estimate of the likely future actions of the attackers. We implemented our approach in a prototype tool, called Nexat, and validated its accuracy leveraging a dataset from a hacking competition. The evaluations show that Nexat is able to predict the next steps of attackers with very high accuracy. In particular, Nexat achieves a 94% accuracy in predicting the next actions of the attackers in our prototype implementation. In addition, Nexat requires little computational resources and can be run in real-time for instant prediction of the attacks.