IoT Protocol-Enabled IDS based on Machine Learning

Abstract

During the last decade, Internet of Things (IoT) devices have become widely used in smart homes, smart cities, factories, and many other areas to facilitate daily activities. As IoT devices are vulnerable to many attacks, especially if they are not frequently updated, Intrusion Detection Systems (IDSs) must be used to defend them. Many existing IDSs focus on specific types of IoT application layer protocols, such as MQTT, CoAP, and HTTP. Additionally, many existing IDSs based on machine learning are inefficient in detecting attacks in IoT applications because they use non-IoT-dedicated datasets. Therefore, there is no comprehensive IDS that can detect intrusions that specifically target IoT devices and their various application layer protocols. This paper proposes a new comprehensive IDS for IoT applications called IP-IDS, which can equivalently detect MQTT, HTTP, and CoAP-directed intrusions with high accuracy. Three different datasets were used to train the model: Bot-IoT, MQTT-IoT-IDS2020, and CoAP-DDoS. The obtained results showed that the proposed model outperformed the existing models trained on the same datasets. Additionally, the proposed DT and LSTM models reached an accuracy of 99.9%.