Predicting Cyberattacks on Connected Healthcare Devices Using LightGBM-based Approach

The rapid digitization of healthcare and the widespread integration of connected medical devices have revolutionized patient care. However, this transformation has significantly increased the attack surface for cyber threats, jeopardizing the confidentiality, availability, and integrity of critical health information and patient safety. Traditional security mechanisms struggle to keep pace with the evolving nature of cyberattacks, necessitating intelligent, efficient, and real-time detection models. In this work, we propose a machine learning-based cybersecurity framework that leverages Random Forest- based feature selection and LightGBM classification to proactively detect and mitigate cyber threats targeting IoMT ecosystems. Our model is designed to identify anomalous behaviors in network traffic, predict attack vectors, and automate intrusion detection. Unlike deep learning-based solutions that are computationally intensive, our approach optimizes feature selection for high accuracy while ensuring real-time applicability in resource-constrained IoMT environments. Experimental results on the CICIoMT2024 dataset demonstrate the high efficiency and robustness of our model, achieving 99% overall accuracy across multiple attack types, including DDoS, DoS, spoofing, and reconnaissance threats. The model executes in just 158.76 seconds, significantly outperforming deep learning models in terms of computational efficiency and real-time adaptability. By providing a fast, scalable, and effective cybersecurity solution, our approach strengthens the cyber resilience of health care infrastructures, ensuring uninterrupted medical service delivery, patient data protection, and enhanced trust in connected healthcare technologies.