VOTING COMBINATION BASED ENSEMBLE OF SOFT COMPUTING TECHNIQUES FOR INLINE INTRUSION DETECTION SYSTEM

The intrusion detection systems are a powerful tool in the organization for keeping its computing resources protected by spotting an intruder’s activities. The intrusion detection system supervises computer systems, looking for signs of unauthorized users or misuse. The researchers are trying to design and implement intrusion detection systems that are easy to use and easy to install with more accuracy. Some existing intrusion detection system is designed and evaluated offline and not suitable for deploying inline intrusion detection purpose. Soft computing techniques are progressively being used for intrusion detection system. In this paper, we present the ensemble approach of different soft computing techniques to design and implement the inline intrusion detection system. Three neural networks based base classifiers are studied and implemented. Neuro-fuzzy neural network, Multilayer Perceptron and Radial Basis Function neural network have been constructed and then combined using voting combination method ensemble. The performances of base classifiers are separately evaluated in term of classification accuracy, false positive rate, false negative rate, sensitivity, specificity and precision. The trained models of these classifiers are combined to construct one ensemble classifier using the voting combination method. The performance of the proposed ensemble classifier is evaluated and compared with the performances of base classifiers. We could show that the proposed ensemble classifier of Neuro-fuzzy, Multilayer Perceptron and Radial Basis Function neural network is superior to the individual base classifier for intrusion detection in terms of classification accuracy and sensitivity. It is also found that the proposed intrusion detection system has reasonable classification accuracy, the best sensitivity and false negative rate with good false positive rate on test data set. The experimental results show that the base classifiers take very less time to build models and the proposed method of ensemble for intrusion detection system takes very less time to test data set. These advantages can help to deploy the intrusion detection system using the proposed method to capture and detect online packets.