A TRUST-BASED ENSEMBLE MACHINE LEARNING FRAMEWORK FOR INTRUSION DETECTION IN MEDICAL INTERNET OF THINGS ENVIRONMENTS

Abstract

The high rate of Internet of Medical Things (IoMT) devices spread in clinical settings has provided critical attack surfaces that cannot be countered by conventional security measures. This paper suggests a trust-based collective machine learning framework of intrusion detecting in healthcare internet of things networks. The framework uses Mutual Information (MI) to generate dimensionality reduction by using 45 input features to generate 34 discriminative features, and then an ensemble classifier is used with hard-voting, comprising K-Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Machine (SVM). The suggested model is tested on the WUSTL-EHMS-2020 data in a real-world healthcare monitoring benchmark with 16,318 network flow and patient biometric records and benchmarked with three individual baseline classifiers. The ensemble had a 95 percent accuracy, specificity of 0.99, and sensitivity (recall) of 0.65 on the minority attack class and AUC of 0.82, which is better than all the individual baselines. The training set was used only and Synthetic Minority Over-Sampling Technique (SMOTE) was used to reduce class imbalance. The findings indicate that the framework in question is capable of making reliable differentiation between malicious and regular data related to network operation, hence justifying the implementation of reliable, secure IoMT systems in mission-critical healthcare settings.

Keywords : Medical Internet of Things (IoMT); Intrusion Detection; Ensemble Learning; Mutual Information; SMOTE; WUSTL-EHMS; Trust-Based Security.