A Neural Network Model for Intrusion Detection Using a Game Theoretic Approach

The problem of intrusion detection in the computer networks is not new and various methodologies have been formulated to address the same. A game-theoretic representation was also formulated, using one of the oldest game playing techniques, the minimax algorithm to solve this problem. It exploited the adversary like situation between the intruder and the Intrusion Detection System (IDS) and the essence of this approach lies in the assumption that the intruder and the IDS have complete knowledge of the network and each other's strategy. The solution for the intrusion detection problem via game theory gives the detection probability by which the IDS can detect the malicious packets on a given network when the probabilities with which the intruder sends the malicious packets on the various paths leading him to the target are known to the IDS. However, in the real world scenario, the role of the intruder and the IDS is dynamic, if the attack is detected or goes undetected the intruder tries to breach the network again with a different approach or the IDS tries to defend the network with a different strategy respectively. The next strategy for either of the two can be learnt by experience and thus, this paper, models an artificial neural network to represent this game-theoretic representation. The modeled neural network gives the detection probability of an attack by the IDS when the probabilities of sending malicious packets on the various paths leading the intruder to the target are given as an input pattern to the neural network.