An ensemble framework of anomaly detection using hybridized feature selection approach (HFSA)

Rapid growth and popularity of internet has re-emphasized the significance of the intrusion detection system in network security. To overcome the vulnerabilities of network security researchers have come up with different frameworks of intrusion detection system using data mining. Feature selection is a significant method to develop a time and cost effective intrusion detection system. The time consumption in building up the classifiers model enhances the efficiency of the system. This work conducted on the analysis of some approaches of intrusion detection using some machine learning methods with wrapper approaches, which is a type of feature selection methodology. Our paper mainly focuses on the classification preciseness of 3 different classifiers using the minimal amount of features selected by three different wrapper search methods on the well-known public type NSL-KDD dataset and showing the comparisons among them. The 3 basic classifiers are Bayesian Network, Naive Bayes and J48. Best First, Genetic Search and Rank Search have been used as the wrapper search methods. The study proposed an ensemble type of a classification model with a hybrid feature selection method based on the research framework. By using the hybrid feature selection method 12 critical features are chosen and with the combination of basic classifiers, a reliable model is developed to differentiate normal and anomaly. Moreover, the result shows a convenient false positive rate of 0.021. Experiment showed that our proposed ensemble approach showed better result than Naive Bayes, Bayesian Network and J48 classifier. Experiments have been conducted on the NSL-KDD dataset using WEKA 3.6 library functions.