A Multi-level Correlation-Based Feature Selection for Intrusion Detection

Intrusions (or threats) have been considerably increased due to the rapid growth in Internet and network technologies. Nowadays, the world is moving more toward a digital world in this era of networks; it makes more vulnerable to attacks. Intrusion detection models have proved to be a robust method in achieving high security in the network. The detection capacity of the intrusion detection model depends on the training set. High-dimensional dataset increases complexities, higher resource utilization, and affects system accuracy. Many researchers have suggested intrusion detection methods with reduced dimensions training set. However, they have not applied the multi-level-based correlation among attributes. This paper analyzed the network data and proposed a multi-level correlation-based feature selection method. It selects significant features and reduces the size of the training set. We have applied a classifier that learns from the training set and detects attacks; the proposed method enhanced the detection capacity. This work provides a detailed analysis of the UNSW-NB'15 dataset with binary classes (normal and attack) and multi-classes (normal and attack categories); it also shows the effectiveness of the UNSW-NB'15 dataset, which maintains a high category. The proposed method is executed on a high-dimensional dataset UNSW-NB'15. Finally, the experimental results are compared with existing techniques that show the better performance of the proposed method.