An Artificial Intelligence–Driven Adaptive Security Architecture for Cyberattack Detection in Internet of Things Networks

Abstract

With the rapidly growing number of IoT devices and their networks, the surface area for cyberattacks has also expanded, and traditional security measures are struggling to mitigate the problem effectively. The inability to protect IoT networks is a nightmare made real by the likes of large-scale botnets. One of the main factors to be considered is the unique characteristics of IoT traffic, such as device heterogeneity, rapid traffic changes, and limited inbuilt security. A group of researchers has presented a brand-new AI-driven intrusion detection system to give a much-needed boost to the security of IoT networks, which has been validated using a real-life IoT benchmark dataset. This N-BaIoT dataset combines regular IoT traffic and Mirai botnet attacks to create a dataset that the framework can train on. 
A supervised learning technique, the researchers use a Random Forest classifier and feed it a huge set of multi-scale statistical traffic features that capture time-related, jitter and communication between hosts details, in order to develop their supervised training dataset of 162,833 traffic instances, each described by 115 characteristics and a single label, 0 or 1. They split the dataset into 80/20 proportions and ran their experiments on the smaller portion, evaluating how well the framework performed using standard metrics such as accuracy, precision, recall, F1-score, and confusion matrix analysis. 
Notably, the results demonstrate that the proposed system can distinguish between clean and malicious IoT traffic with no false alarms and no missed attacks, indicating that the statistical patterns employed by the system and the ensemble learning technique have their strengths in IoT intrusion detection.