Attribute Selection and Ensemble Classifier based Novel Approach to Intrusion Detection System
Kunal, Mohit Dua
Abstract
Abstract The rapid expansion of computer networks is causing vulnerabilities to occur, which in turn is compromising security. Thus, there is a need to monitor network traffic transmitted over these networks. Network vulnerabilities can cause huge loss to organizations. Hence, there is a need of effective and robust Intrusion Detection System (IDS) that can detect anomalies and raise alarm if such activity occurs. Recently, machine learning techniques have gained popularity for building intelligent Anomaly Detection Systems to detect novel attacks. These techniques produce diverse levels of accuracy and precision. Machine learning classifiers are first trained using training data to learn attack pattern of intrusions and then, tested against test data to differentiate normal data and anomalous data. The proposed work in this paper uses ranker-based attribute evaluation technique to reduce number of attributes and evaluates the implemented model using ensemble of IBk(K-NN), Random Tree, REP Tree, j48graft and Random Forest classifiers. The obtained experimental results show accuracy of 99.72% for binary classification and 99.68% for multi-class classification. The model uses NSL-KDD dataset for evaluating the performances. It is revealed by the experimental results that time complexity and computational expense are minimized due to dimensionality reduction of data and hybrid approach that is used for classification.