Anomaly Detection on Clinical IOT and Embedded Devices using Deep Learning

Abstract

The sudden increase in the adoption of the Internet of Medical Things (IoMT) has greatly improved the quality of healthcare delivery; however, it has also introduced new cybersecurity threats capable of jeopardizing patient safety. Conventional intrusion detection systems and traditional machine learning (ML) techniques face significant limitations in IoMT environments due to their high computational requirements and challenges associated with class imbalance. This study developed an optimized deep learning-based intrusion detection model to improve anomaly detection in resource-constrained IoMT networks using the CICIoMT2024 dataset. The study explored the implementation of an efficient separable One-Dimensional Convolutional Neural Network (1D-CNN) combined with focal loss to address the issue of class imbalance across the 19-class attack taxonomy. The results showed that the proposed separable 1D-CNN achieved a high classification accuracy of 98.54% while maintaining minimal computational resource usage. In addition, the detection of minority attacks, such as spoofing and reconnaissance port scanning, achieved a recall of 64% and an F1-score of 91% through the use of focal loss. The study concluded that lightweight deep learning architectures enhanced with separable convolutions and focal loss can provide an effective and scalable solution for securing IoMT devices and safeguarding patients from cyber threats.