A Proposed Intrusion Detection System Based on an Improved Random Forest Using a Double Feature Selection Method

Cyber-attacks today are a source of great concern due to the increase in the use of the Internet in many areas, which has allowed increasing intrusion on networks and attempts to damage systems and others. Therefore, to stay up with the evolution of cyber-attacks, intrusion detection systems must be constantly improved. Intrusion detection system is a technique that may be applied to track both known and unidentified breaches before one of them damages network hardware. One of the very important things that has a big role in the strength of the system is the selection of good features in training the system. In this research paper, intrusion detection systems are proposed based on reducing and selecting features through the use of a “double feature selection” with the random forest algorithm. Experiments were performed on a data set NSL-KDD (it dataset from the Canadian Institute for Cybersecurity). By evaluating the performance, a system accuracy of 0.9981, a training time of 3.47 sec and a detection time of 0.24 sec were obtained. The proposed work was compared with related work using the same algorithm and dataset. The system proved superior to many of the proposed systems in terms of accuracy of the system, recall, precision, the time spent in training the system and the time of detection.