Dynamic Improved Hybrid Fuzzy Jordan Network For Robust And Efficient Intrusion Detection System

The security is an important aspect to protect the infor- mation systems and networks against attacks. Intrusion Detection System assists information systems practice to deal with attacks. A hybrid fuzzy Jordan artificial neural network was proposed to detect intrusion based on learned model. However fuzzy Jordan artificial network is static which cannot handle the dynamic behavior of in- truder, the behavior-based intrusion detection techniques detects intrusion by observing a deviation from normal or expected behavior of the system or the users. The model of normal or valid behavior is extracted from reference information such as timestamps, sizes of the payloads, and TCP-flags of all packets etc and is collected by vari- ous means which can be incorporated with other static feature to improve the accuracy of hybrid fuzzy Jordan artificial neural network. The time varying features are extracted from captured packet information with sliced time window. The first order statistics Variance, Skewness and Kurtosis, second order statistics correlation, entropy and moment of time varying features are calculated to use in the classification along with actual value of this features to improve the prediction accuracy. The network struc- ture of hybrid fuzzy Jordan artificial neural network is de- cided dynamically based on the error achieved while learn- ing stage.