A Balanced and Feature-Optimized Intrusion Detection Approach for Threat Mitigation in IoMT

The expanding usage of Internet of Medical Things devices has raised considerable security issues, especially integrity breaches that compromise data trustworthiness and patient safety. Conventional Intrusion Detection Systems (IDS) frequently face difficulty with imbalanced datasets, resulting in skewed classification performance. Moreover, high-dimensional feature sets elevate computational complexity, complicating real-time attack detection. The proposed work research tackles these problems by including Synthetic Minority Over-sampling Technique for balancing datasets and Feature Selection Techniques to boost classification effectiveness. The recommended method is assessed with five machine learning algorithms: Decision Tree, Random Forest, XGBoost, Adaptive Boosting, and K-Nearest Neighbors. Study reveal that SMOTE remarkably uplift recall scores for underrepresented attack classes, enhancing the detection ability of integrity attacks within IoMT networks. Feature selection enhances the model by eliminating unnecessary attributes while preserving strong classification accuracy. Among the assessed models, DT and RF attained peak performance with an F1-score of 1.00 both prior to and following dataset balancing. XGB and KNN also demonstrated steady performance, whereas ADB presented enhancements after balancing but had a marginally reduced recall for some attack categories. The research shows that integrating data balancing and feature selection improves IDS effectiveness, resulting in greater efficiency and reliability for IoMT security purposes.