MACHINE LEARNING TECHNIQUES TO DETECT ROUTING ATTACKS IN RPL BASED INTERNET OF THINGS NETWORKS

The Internet of Things (IoT) can be considered an interconnection of heterogeneous smart devices via the internet to exchange information among the devices for control and operating. They are vulnerable to numerous routing attacks due to their open nature, worldwide connection, and resource-restricted nature of smart devices and wireless networks. RPL (Routing Protocol for Low Power Lossy Networks) is a widely used routing protocol in IoT-based networks for establishing routing paths for resourceconstrained devices. RPL's built-in security features are ineffective at preventing most routing attacks. We present machine learning approaches to identify threats in RPLbased IoT networks in this paper. For this implementation, we used the synthetic data set created using the NetSim software, which contains the traffic traces of Sinkhole, Blackhole, Sybil, Selective Forwarding, DIS flooding and DIO suppression attacks with 21 features and two labels as normal and attack. We implement seven types of machine learning algorithms like K-Nearest Neighbors (KNN), Logistic Regression (LR), Random Forest (RF), Gaussian Naive Bayes (GNB), Decision Trees (DT), AdaBoost (AdB) and Multilayer perceptron (MLP). All the algorithms performance was evaluated using the performance metrics like Accuracy, Precision, Recall, F1-score and AUC for four test data hold out conditions like 10%, 20%, 30% and 40%, respectively. The DT classifier achieved the highest accuracy of 92.6% for 10% test data hold out, and at the same time, it scored the highest precision and F1-score of 0.946 and 0.955,respectively, compared to all other classifiers for different test data hold out conditions. The classifiers LR, GNB and MLP have the maximum Recall value of 1.00 for all the states. DT classifier performance was highest among all the remaining algorithms compared to all the seven machine learning algorithms. For AUC metrics, RF classifiers got the more AUC of 0.946 for 20 % hold out data, and the GNB classifier got the less AUC of 0.623 for 10% hold out data compared to all other classifiers.