Dynamic learning framework for IoT intrusion detection using statistical approach and unsupervised learning.
Mohamed Khalafalla Hassan, Sharifah H. S. Ariffin, Mutaz Hamad, Safa Elhadi, Bushra Mohammed Ali Abdalla · 2025
The Internet of Things (IoT) is increasingly becoming integral in various sectors like transportation and healthcare, driving the development of new services. This paper proposes an innovative security approach for IoT, utilizing feature selection, dynamic learning with statistical change detection, and Automated Machine Learning (AutoML) for ongoing model refinement. Applied to a comprehensive IoT dataset, this method effectively tackles feature drift in dynamic environments, underscoring the need for flexible cybersecurity tactics. It demonstrates the proposed framework’s role in transforming attack detection and classification in IoT. The testing involved 33 attacks on an IoT network with 105 devices. The results indicate that our methodology significantly improves classification performance by 8% to 67%, depending on the drift percentage.