Enhancing the Accuracy of Intrusion Detection Systems by Reducing the Rates of False Positives and False Negatives Through Multi-objective Optimization

Intrusion detection systems (IDSs) are the fundamental parts of any network security infrastructure given their role as layers of defense against hackers. However, IDSs generate frequent instances of false alerts and miss a lot of real attacks that block the normal traffic and threaten the network security. It is not possible to identify a missed intrusion using one IDS, so multiple IDSs are used since they respond differently to the same packet trace and produce different alert sets. Actually, an attack missed by an IDS can be detected by another while inspecting the same network traffic. In this paper, we propose a multi-objective optimization process that aims to identify false negatives and false positives from the sets of alerts generated by multiple IDSs. In the first step, low-level alerts are clustered into meta-alerts to give a better understanding of the output of each IDS. Then, a filtering step is performed having as input the distinct meta-alert sets generated by different IDSs and as output the set of potential false negatives collecting the meta-alerts detected by some IDSs and missed by others. Meta-alerts generated by all IDSs are discarded since they cannot be missed attacks. Later, a clustering inter-IDS step is performed to group together similar meta-alerts generated by different IDSs. This clustering step aims to avoid the redundancy between the alerts generated by more than one IDS. Finally, a binary multi-objective optimization problem is used to detect false negatives and false positives. The proposed method is evaluated using a real network traffic, DARPA 1999 and NSL-KDD data sets. Experimental results show that the proposed process outperforms concurrent methods for false negatives and false positives detection.