Network Intrusion Detection using Machine Learning and Voting techniques

Abstract

As the result of recent advent and rapid growth of the Internet, there have been an increasing number of corporations relying on computers and networks for communications and critical business transactions. Because of the network complexity and advanced hacking techniques, such reliance on computer networks often presents unanticipated risks and vulnerabilities. A huge volume of attacks on major sites and networks have been recently reported including those of private companies, government agencies and even military classified networks. Therefore, it is important to deploy protection measures for networks and their services from unauthorized modification, destruction, or disclosure of sensitive information. Intrusion detection systems (IDS) have emerged as an important part of today's network security infrastructure which can monitor the network traffic and detect possible attacks. Currently existing IDS suffer from low detection accuracy and system robustness for new and rare security breaches. To improve detection capability of IDS, this chapter proposes an innovative Machine Learning (ML) framework in which different types of intrusions will be detected with different classifiers, including different attribute selections and learning algorithms. Outputs of these classifiers are then combined by appropriate voting techniques. Experiments on the KDD-99 dataset show that our approach obtains superior performance in comparison with other state-of-the-art detection methods, achieving low learning bias and improved generalization at an affordable computational cost.