Applied Machine Learning for Securing the Internet of Medical Things in Healthcare

Abstract

Integrating mobile computing technologies and human health activities using IoMT devices can accelerate biomedical discovery and improve the efficiency of healthcare research and delivery. However, the quality of the collected health data is critical for the success of these efforts. In this paper, a hybrid intrusion detection system is proposed to identify cyberattacks in real time on medical devices. The system combines a logistic regression-based detector using network traffic features with a gradient-boosted tree-based detector using medical sensor features. Evaluation of the system using a publicly available dataset shows an accuracy score of 95.4% using only 11 features, compared to the current best accuracy of 92.98% achieved by artificial neural networks using 40 features. Additionally, by combining the decisions of the two individual detection systems, the number of attacks detected is increased from 111 to 305 out of a total of 423 attack instances, improving the sensitivity score and addressing the challenge of effectively and efficiently integrating different detection technologies in a hybrid intrusion detection system. To the best of the authors’ knowledge, this is the first attempt to combine multiple misuse detection models in a hybrid system to secure both IoMT devices and their networking equipment across the entire healthcare spectrum of the IoMT.