Abstract: Cyber security has become an increasingly important area in computer science in response to the expansion of private sensitive information. Intrusion can be defined as an uncertified access, which aims to compromise integrity, confidentiality and availability of data. Conventional intrusion prevention method such as access control firewalls and encryption cannot fully prevent system from advanced attacks. Intrusion Detection System has become a crucial part of computer security, which is used in detecting the above-mentioned threat.This paper presents an agent based Anomaly intrusion detection and prevention system using Reinforcement Learning Technique. The system uses two agents, the first agent attacks the network system while the second agent detects the attack and classify it to be either normal, dos, probe, u2l and u2r attack, the orange line represents the reward receive by the attacking agent while the blue line represents the reward of the agent detecting and classifying the attack. The attacking agent receives a total reward 5 while the defending agent received a total reward of 95. This means that the defending agents performs more better in detecting and classifying attacks that is being carried out by attacking agent. The diagram also shows the loss values of the both agent during training. The both agent has a loss value below 0.5 during training. Figure 5 shows the performance of the defending agent in classifying an attack currently. The agent obtained individual accuracy in each of the attack. The accuracy are as follows, normal 0.79%, DoS 0.94%, R2L 088%, Probe 0.94% and U2R 0.99%.
Keywords: Reinforcement Learning, Deep Q-learning Network, Intrusion detection, Anomaly attack.
| DOI: 10.17148/IJARCCE.2021.10114