Abstract

With the development of the Internet of Medical Things (IoMT), the process of exchanging and monitoring patients improved through connected devices. However, this connectivity brings a lot of cybersecurity risks; therefore, there is a need for strong intrusion detection systems (IDS). In this study, we propose a hybrid IDS model that accomplishes malicious activities detection based on a fusion of machine learning (Random Forest) and deep learning (Long Short-Term Memory). For model training and evaluation, we made use of the MedBIOT dataset consisting of attack and normal traffic samples. To capture different traffic patterns, diverse feature selection techniques were used, such as mutual information and Hurst exponent metrics. The combination of LSTM’s temporal pattern recognition and Random Forest’s predictive power synergizes to create a new, fundamental hybrid model. We show that the hybrid model yields better precision, recall, and accuracy than Random Forest or LSTM models alone. This hybrid approach addresses the security challenges related to IoMTs and offers a way to secure a critical medical network in a scalable and efficient manner. Finally, the study also demonstrates the feasibility of using machine learning and deep learning to develop adaptive and robust IDS frameworks for the IoMT.