Machine Learning-based Multiple Attack Detection in RPL over IoT

By connecting the low-power smart embedded devices through the internet, the Internet of Things (IoT) is a prominent technology in smart world construction. The IoT has numerous smart applications ranging from simple home automation to a complex surveillance system. However, IoT poses several security issues due to its heterogeneity and ad hoc nature. It is crucial to detect IoT security threats using appropriate mechanisms. It is prominent to detect the attacks earlier since IoT devices have a low storage capacity, and conventional high-end security solutions are not appropriate for IoT. It implies that an intelligent security solution like machine learning solutions has to be designed for IoT. Although several works have been proposed using Machine Learning (ML) solutions in attack detection, little attention has been given to IoT networks. To detect attacks before making a huge impact on the IoT, this paper proposes a novel Machine Learning based secure RPL routing (MLRP) protocol. Initially, the MLRP protocol creates a complex dataset, including normal and attack behavior using the Cooja simulator. Secondly, the dataset is learned by the machine to efficiently detect the attack behaviors that are version, rank, and Denial of Service (DoS). Moreover, the MRLP classifies the attack types using the Support Vector Machine (SVM) classifier. To improve the performance of SVM, improved Principal Component Analysis (PCA) is introduced to reduce the dimensionality of the dataset effectively. Finally, the simulation results demonstrate that the proposed MLRP protocol increases the attack detection accuracy with precision and fits the IoT environment. The MLM-RPL attains 76.8% of PDR by using only 1474 control packets over a 30 node IoT scenario.