Semi-Supervised Network Intrusion Detection with Two Phased Hybrid Ensemble Method

The use of network connected devices has grown exponentially in recent years. The emergence of Internet of Things (IoT) with cloud computing has been a major catalyst for this growth with more devices getting connected with wireless networks and making smart decisions with less human interaction. As such, security concerns of these network devices have also gained significant attention especially due to new and unknown attacks occurring more frequently with increased sophistication. To detect such attacks, an Intrusion Detection System (IDS) has become a vital component in network applications. However, network devices produce large volumes of high dimensional data which makes feature selection process of an Intrusion Detection System (IDS) a challenging task. Furthermore, detecting new and unknown attacks are becoming increasingly harder because the attacks themselves are becoming more sophisticated with no previously known attack signatures. This study proposes a novel semi-supervised two-phased hybrid ensemble method for muti-class classification capable of detecting new and unknown attacks. Each learner within the ensemble is constructed using a two-phased classification and voting mechanism where the first phase is constructed with a set of binary classifiers transformed from an adaptation of the One-vs-One method, and the second phase is constructed with a set of multi-class classifiers from combinations of all attack classes. Finally, the proposed method was tested on both wired and wireless applications using two well referenced datasets NSL-KDD and AWID, and the results were compared against other leading studies in literature. While it was able to achieve a higher detection rates in both applications compared to other existing studies, the notable result was in the wireless application where it was able to achieve a remarkably exceptionally higher detection rate of 92% which is a 7% improvement over the current leading study.