An Experimental Analysis of Hybrid Classification Approach for Intrusion Detection

Background: Recently network security is achieved using intrusion detection, in which data mining techniquesare used as a new methodology. The vital features considered is one of the major aspects that affect the efficiency of the Intrusion Detection System (IDS). Methods: The key idea of this work is to propose a feature selection method to discover useful features and to classify user behaviour patterns of system features from the network traffic data using classification approaches. In the process of selecting significant features, the dimensions of data is reduced and the features are sorted by finding the accuracy of each attribute and then selects the best vital features among them based on its accuracy value. Also this work aims to choose a hybrid classifier model (ABC-SVM) based on Artificial Bee Colony (ABC) and Support Vector Machine (SVM) algorithms to construct a perfect IDS using KDDCup'99 dataset. Results: The result analysis indicates that the features selected improve the accuracy rate of ABC-SVM than using all features. Also the hybrid algorithm is better than other traditional algorithms with respect to the performance measures such as detection rate, specificity and training time.