Network Intrusion Detection with Two-Phased Hybrid Ensemble Learning and Automatic Feature Selection

The use of network connected devices has grown exponentially in recent years revolutionizing our daily lives. However, it has also attracted the attention of cybercriminals making the attacks targeted towards these devices increase not only in numbers but also in sophistication. To detect such attacks, a Network Intrusion Detection System (NIDS) has become a vital component in network applications. However, network devices produce large scale high-dimensional data which makes it difficult to accurately detect various known and unknown attacks. Moreover, the complex nature of network data makes the feature selection process of a NIDS a challenging task. In this study, we propose a machine learning based NIDS with Two-phased Hybrid Ensemble learning and Automatic Feature Selection. The proposed framework leverages four different machine learning classifiers to perform automatic feature selection based on their ability to detect the most significant features. The two-phased hybrid ensemble learning algorithm consists of two learning phases, with the first phase constructed using classifiers built from an adaptation of the One-vs-One framework, and the second phase constructed using classifiers built from combinations of attack classes. The proposed framework was evaluated on two well-referenced datasets for both wired and wireless applications, and the results demonstrate that the two-phased ensemble learning framework combined with the automatic feature selection engine has superior attack detection capability compared to other similar studies found in the literature.