Unsupervised Representation Learning Approach for Intrusion Detection in the Industrial Internet of Things Network Environment

The large-scale evolution of Internet and devices connected to the internet have led to various companies and organizations to protect their data on the internet to implement large scale IoT networks such as IIoT in the industrial point of view. Such large-scale networks need to be protected from malicious attacks. This makes it crucial for the need of an intrusion detection system that can protect the privacy and the data in an IoT network and keep the network secure. Most of the existing works are based on a supervised approach where the data in expected to be labelled and use complex deep learning architectures. In our research we propose an unsupervised intrusion detection model that was implemented using the FUZZY C Means algorithms using autoencoders that provide the best detections of the intrusions into the networks. Various other models like the Gaussian-Mixture Model, K means and the HMM have also been used to develop an unsupervised intrusion detection system. The WUSTL_IIOT_2021 and the OPCUA datasets has been used to compliment the effectiveness of our algorithms and to demonstrate the need for more unsupervised approaches for IDS. By our proposed method we have obtained a maximum accuracy of 97% on the Fuzzy-C Means approach and 95% on GMM, HMM and K-Means. Our proposed approach is well in competition with the existing IDS using various complex supervised techniques. These results are superior to the existing frameworks as the system does not expect the data to be labelled as it would mean that the system already know about the features that would cause an attack which is not expected in practical conditions.