Protocol-Specific and Sensor Network-Inherited Attack Detection in IoT Using Machine Learning

Abstract

For networks with limited resources, such as IoT-enabled smart homes, smart industrial equipment, and urban infrastructures, the Routing Protocol for Low-power and Lossy Networks (RPL) was developed. Additionally, a number of optimizations have been suggested for its application in other contexts, such as smart hospitals, etc. Although these networks offer efficient routing, the lack of active security features in RPL makes them vulnerable to attacks. The types of attacks include protocol-specific ones and those inherited by wireless sensor networks. They have been addressed by a number of different proposals, many of which have achieved substantial prominence. However, concurrent handling of both types of attacks is not considered while developing a machine-learning-based attack detection model. Therefore, the ProSenAD model is proposed for addressing the identified gap. Multiclass classification has been used to optimize the light gradient boosting machine model for the detection of protocol-specific rank attacks and sensor network-inherited wormhole attacks. The proposed model is evaluated in two different scenarios considering the number of attacks and the benchmarks for comparison in each scenario. The evaluation results demonstrate that the proposed model outperforms with respect to the metrics including accuracy, precision, recall, Cohen’s Kappa, cross entropy, and the Matthews correlation coefficient.