CYBERSECURITY IN SMART HEALTHCARE: A MACHINE LEARNING APPROACH

Abstract

The adoption of Internet of Things (IoT) technologies in medical devices has greatly enhanced healthcare capabilities. This enables continuous patient monitoring, real-time diagnostics, and remote care. However, this connectivity also introduces significant cybersecurity threats that can compromise patient safety and system integrity. This paper presents a machine learning-based framework for detecting threats in IoT-enabled medical devices. This study utilizing the WUSTL-EHMS-2020 dataset that taking a collection of network traffic from real-world healthcare IoT systems. A comparative evaluation of multiple classifiers was conducted to assess detection effectiveness and computational efficiency. In terms of accuracy value, the Decision Tree (DT) achieves highest value of 0.97. The Random Forest (RF) model demonstrated more optimum performance across metrics with accuracy at 0.94, precision of 0.95, recall of 0.56, and F1-score of 0.70. Meanwhile, XGBoost (XGB) achieved the highest Area Under the Curve (AUC) score at 0.95, indicating strong overall classification performance. Conversely, Gaussian Naive Bayes (GNB) exhibited the weakest results, with an accuracy of 0.86, F1-score of 0.46, and the lowest AUC score of 0.73. Notably, K-Nearest Neighbors (KNN) achieved the fastest training time of just 0.001 seconds, offering a preferable option for deployment in time-sensitive environments. These results highlight the trade-offs between accuracy, speed, and robustness in machine learning-based intrusion detection systems. This study underscores the potential of intelligent threat detection models in strengthening the security of modern medical IoT infrastructures, all while balancing computational constraints.