Research Authors & Affiliations

Abdullah Rakib Akand¹ & Md. Sazzadur Rahman²

¹Dept. of Computer Science and Engineering, Asian University of Bangladesh, Dhaka, Bangladesh
²Institute of Information Technology, Jahangirnagar University, Savar, Dhaka, Bangladesh

Conference Information

Published in: 2026 IEEE QPAIN
Publisher: IEEE | Location: Chittagong, Bangladesh
Conference Date: 16-18 April 2026
Added to IEEE Xplore: 11 June 2026
DOI: 10.1109/QPAIN69676.2026.11546448

The rapid proliferation of Industrial Internet of Things (IIoT) devices has introduced significant security vulnerabilities, particularly from zero-day attacks that traditional signature-based systems fail to detect. This paper proposes a novel unsupervised ensemble method, the Integrated Multi-Modal Ensemble (IMME), for robust zero-day attack detection in IIoT environments. The ensemble integrates five distinct anomaly detectors: Adaptive Streaming Clustering, Autoencoder (AE) reconstruction error, Mahalanobis distance in latent space, Isolation Forest (IF), and an O(1) Sliding-Window Rarity Kernel. Evaluation using the CICIoT 2023 dataset demonstrates that IMME achieves a high ROC-AUC of 0.9560, an Unknown Detection Rate (UDR) of 47 %, and an Open-Set Classification Rate (OSCR) of 47.5 %, while maintaining 100 % precision through a validation-optimized thresholding approach. The system exhibits a throughput of 79,000 flows/sec, proving its viability for real-time IIoT deployment.