A TWO-TIER INTRUSION DETECTION SYSTEM FOR IOT DEVICES USING MACHINE LEARNING

Abstract

IoT devices are bringing more and more benefits to human, leading to a significant increase of their productions to supply demands of the devices. Due to the widespread productions and lack of security standards, IoT devices become targets of malicious activities such as intrusions and hacking. In order to address these problems, this thesis introduces a two-tier intrusion detection system for IoT devices which applies machine learning algorithms. The proposed system consists of two tiers. The first tier is a binary instruction detection model implemented in the gateways of IoT networks, while the second tier is a multi-class classification model of malcious activites located on a remote cloud server. In order to detect intrusions, the system needs to capture and extract the attributes of packets such as protocol, port, total duration, and then inputs the data simultaneously into two models to perform detections. In this thesis, we develop proposed models using the UNSWNB15 dataset focusing on 9 common types of attacks. The evaluation of models shows the accuracy of over 95%, while efficiently detecting malicious activities experimental environment