Intrusion detection using fuzzy logic and evolutionary algorithm techniques

An Intrusion Detection System should optimally be capable of detecting both known attacks (misuse detection) and unknown attacks (anomaly detection combined with non-self classification). This thesis research studies the problem of automating the generation of a high-fidelity ‘detection model’ that can recognize both known and variations on known attacks through the use of a Fuzzy Learning Classifier System. Experimental results on the classic KDDCup’99 benchmark dataset reveal that the proposed model outperforms published results obtained with the well-known C4.5 classification program. Fuzzy Logic and Evolutionary Computation are very robust in modeling real-world problems like intrusion detection. Therefore, the proposed model is aimed at using fuzzy rules for effective intrusion detection with the goal of evolving the rules over time with a Learning Classifier System. This approach is complemented with the optimization of the membership functions for the fuzzy rules using Evolutionary Algorithms. This hybrid approach was shown to significantly improve the accuracy of an Intrusion Detection System --Abstract, page iii