Novel Anomaly Intrusion Detection Using Neuro-Fuzzy Inference System

Conventional approaches to intrusion detection system pose a myriad of problems that exhibit serious impediments to the degree of configurability, extensibility, and effectiveness of the systems. The proposed methodology is a combination of three techniques comprising two machine-learning paradigms. K-Means Clustering, Fuzzy Logics and Neural Network techniques deployed to configure an effective intrusion detection system. Out of the several problems in the traditional techniques of Intrusion Detection Systems, the presence of high rate of false alerts causes unnecessary interference of human analyst. The human analysts in turn perform an intensive analysis repeatedly to distinguish the nature of such alerts and initiate sufficient actions. The approach proposed reveals the advantage of converging K-Means - Fuzzy - Neuro techniques to eliminate the preventable interference of human analyst in such occasions. The technique was tested using multitude of background knowledge sets in DARPA network traffic datasets. The experimental results render remarkable improvement in reducing the false alarms in addition to increased ability to capture intrusion packets that are no similar to the ones in the training datasets.