Trust Management for Deep Autoencoder based Anomaly Detection in Social IoT

Abstract

Social IoT has gained huge traction with the advent of 5G and beyond communication. In this connected world of devices, the trust management is crucial for protecting the data. There are many attacks, while DDOS is the most prevalent BotNet attack. The infected devices earnestly require anomaly detection to learn and curb the malwares soon. This paper considers 9 IoT devices deployed in a Social IoT environment.We introduce a couple of attacks like Bash lite and Mirai by compromising a network node. We then look for traces of malicious behavior using AI algorithms. The investigation starts from a simple network approach - Multi-Layer Perceptron (MLP) then proceeds to ML - Random Forest (RF). While MLP detected the malicious node with an accuracy of 89.39%, RF proved 90.0% accurate. Motivated by the results, the Deep learning approach - Deep autoencoder was employed and found to be more accurate than MLP and RF. The results are encouraging and verified for scalability, efficiency, and reliability.