Intrusion Detection System for Internet of Things Using Image Classification
Mohamed Selim Korium, Jules Merlin Moualeu, Mehar Ullah, Arun Narayanan, Pedro H. J. Nardelli · 2025
The Internet of Things (IoT) is a fast-moving technology that is gradually being integrated into our daily lives. As communication protocols and network technologies evolve, the vulnerability of IoT devices to cyberattacks also increases, fueling the need to address this pressing problem. In this work, we propose an intrusion detection system based on a residual neural network with inductive transfer learning. This learning approach is designed to detect cyberattacks on IoT devices by visually encoding the CIC-IoT-2023 dataset from multivariate numerical data to visual formats (images). Extensive numerical experiments are carried out using the well-known dataset CIC-IoT-2023, which consists of 34 classes. Furthermore, the ensuing results demonstrate the effectiveness of our proposed solution, which achieves an accuracy of 99.35% with a latency of 70.9 ms, a detection time of 99.6 s for the entire dataset, and executes 316.82 predictions per second, outperforming existing solutions in terms of the ability to distinguish between the 34 classes of IoT cyberattacks while reducing overfitting.