Adaptive Memory Graph Transformer-Based Intrusion Detection for Secure Industrial Internet of Things

The rapid growth of Industrial Internet of Things (IIoT) devices has brought with it a heightened threat of cyberattacks in today's networks. In traditional IDS systems, they may not be effective in detecting new and emerging attacks, which hampers their effectiveness in real-world scenarios. To tackle this problem, this paper presents an Adaptive Memory Graph Transformer Intrusion Detection System (AMGT-IDS) based on the CIC IIoT Dataset 2025. The proposed model preprocesses the network traffic data and creates dynamic traffic graph representations to show the communication between IIoT devices. Important structural features are captured using a Graph Attention Network (GATv2) and traffic patterns and attack behavior are learned over time through a Protocol-Aware Transformer. This is then enhanced through the use of an Adaptive Memory Module to detect known and unknown attacks. The proposed model is evaluated using Accuracy(A), Precision(P), Recall(R), F1-Score(F1-S), MCC and AUROC. The experimental results indicate good performance (A of 99.2%, P of 99.0%, R of 98.9%, (F1-S) of 98.9%, MCC of 0.98, and AUROC of 0.99). The results confirm that proposed AMGT-IDS model can successfully detect cyber threats and offer reliable security solution to next generation IIoT network.