Detecting malicious IoT traffic using Machine Learning techniques

Internet of Things (IoT) generates huge amount of data, that needs to communicate between the IoT enabled devices. These communications are vulnerable to security attacks and are malicious enough to cause harm to connected devices. The invasive communication and security breaches have to be identified and should be dealt with in order not to cause further damage and consequences. The objective of this work is to distinguish intentional communications from insecure communications between the IoT devices. The intentional communications can be different from the insecure communications in their patterns. Artificial intelligencebased machine learning approaches have the technologies to identify patterns of the intentional or insecure communications. In this paper, Random Forest, Decision Tree, SVM and 1DCNN have been used to discriminate patterns belonging to intended and unintended messages. To evaluate this technique, IoT-23 dataset is used, the proposed machine learning based approach obtaining a performance of 99.25% accuracy with the benchmark dataset. The proposed approach is compared with the state-of-the-art methods. It is observed that the proposed Random Forest method outperforms the existing ones with sufficient patterns to identify. To enhance the performance of the poorly performing classifiers on the imbalanced dataset, a potential solution to be applied on this dataset is also explored and proposed.