A Comparative Analysis of Network Intrusion Detection System for IoT Using Machine Learning

The size of Internet of Things (IoT) networks is increasing exponentially, and parallelly, the security threats are also escalating. As many of the IoT devices run on limited resources, any intrusion attack based on DoS, packet flooding, man-in-middle and probing is effective to disturb, distract and defunct the networks. A intrusion detection is always a challenging task for the network administrator or any automated software system. A machine learning network-based intrusion detection system (IDS) works efficiently and detects such attacks in any type of networks. It analyzes the packets transmitting over the networks without bothering the IoT devices. Hence, IDS systems are highly crucial and important for IoT network security. This paper proposes a machine learning network-based IDS for securing IoT networks. The proposed technique uses classification techniques to classify a network packet as normal or some kind of malicious attacks. The model was trained with a dataset which is a network logs collected from a network transmitting data from NodeMCU with ESP8266 wi-fi module to a server. The data was captured from the ultrasonic sensor with Arduino and NodeMCU used to monitor a network. For choosing the best detection model out of eight classification based models were studied. The decision tree and random forest are most accurate models as compared to other classification techniques. The comparative analysis of these models is analyzed and discussed in the result section.