Intrusion Detection System for RPL-Based IoT Networks: Mixed Methodology of Graph Neural Networks and Autoencoders

The Internet of Things is happening to everyday objects, but many adopted devices are at risk for continuing security breaches due to limited Availability, Processing, and storage and restrictive communication capability. Below, a number of security issues associated with RPL, which might be an IoT technology, are discussed. Many researches have been made especially on physical assaults and various extraneous traffic can overwhelm the network in a given manner.are used for detection. This work presents an IDS system that uses Flooding, Black Hole, DODAG Version Numbers, and Reduced Rank attacks of the ROUT-4-2023 dataset. Each of the four types of attacks described above is examined using the statistical information graphs of network traffic data. To measure the performance, several machine learning and deep learning models are tested, focusing on both time complexity as well as confusion matrix results. The results suggest that the Transformer model achieves 97% F1- Score, and the Random Forest classifier shows a 99% accuracy level. Further, the Transformer takes only 16.8 mins for the training session at five epochs, demonstrating high accurate and improved training time.