Abstract
The lack of security techniques that assures the integrity of data generated by the Internet of Things networks is one of the major stumbling blocks that hinders the whole development of these networks. As smart devices generate and transmit confidential and sensitive data, in these cases, data leakage can lead in many cases to drastic consequences. These types of attacks are called compromised attacks and the U2R and R2L subcategories are typical examples. Despite various intrusion detection systems designed for detecting compromised attacks, many of them are deficient and suboptimum due to the highly sophisticated behavior of these attacks that mimic normal ones, as well as the rarity of occurrence, and lack of training records keep this area inconclusive. Therefore, this paper presents an ensemble-based intrusion detection system to identify rare hard-to-detect attacks (U2R and R2L) in IoT networks. The ensemble is composed of two main tiers. Each tier consists of a major improved KNN classifier driven by multiple auxiliary classical KNN classifiers. the combined responses of these detection tires are fused to provide the optimal detection performance for minority attacks. The experimental analysis unveiled that the proposed system achieved a high detection accuracy reaches up to 96.65% and 100% for R2L and U2R respectively along with Cohen's kappa coefficient reaches up to 0.9778 and 0.996, which confirms the reliability and robustness of the proposed system to be deployed in loT networks.