Comparison of supervised learning and reinforcement learning in intrusion domain

Abstract

In modern world use of network is increasing exponentially. Network security needs attention of computer science researchers. Intrusion Detection System is software / hardware which detects intruder in the network or host system. Classification plays an important role in Intrusion Detection System. Detection of anomaly or normal traffic is main working philosophy for such type of system. For detection of online traffic, learning of the system is required. In our paper, performance of Supervised Learning and Reinforcement Learning is compared in Intrusion Domain. NSL-KDD data is considered for our work. In that dataset for each object 41 conditional attributes and one decision class attribute are mentioned. Out of 41 attributes, 7 attributes are discrete and 34 attributes are continuous. Using feature ranking method, number of discrete attributes are reduced and these reduced number of attributes are used for classification in Supervised Learning. Some Supervised Learning like CS-MC4, Decision List, ID3, Naive Bayes, C4.5, Rnd Tree are applied on this data set and compared this classification result with classification accuracy derived from Reinforcement Learning combined with Rough Set Theory classifier. © 2012 Springer-Verlag.