A Hybrid IG-PCA and Machine Learning Approach for Accurate Intrusion Detection in IoMT with Imbalanced Data

Abstract

The rapid growth of the Internet of Medical Things (IoMT) has introduced critical cybersecurity challenges, highlighting the need for robust and accurate intrusion detection systems (IDS). This study presents a hybrid machine learning (ML) framework to strengthen intrusion detection in IoMT networks using the CIC-IoMT2024 dataset. The framework combines Information Gain (IG) and Principal Component Analysis (PCA) for feature selection and dimensionality reduction, while SMOTEENN and SMOTETomek are applied to address severe class imbalance. The processed data are classified using Random Forest (RF), K-Nearest Neighbors (KNN), XGBoost (XGB), Multi-Layer Perceptron (MLPC), and Logistic Regression (LR), with hyperparameters optimized through Bayesian Optimization. Performance is evaluated using Accuracy, Precision, Recall, F1-Score, and AUC. Experimental results reveal that the optimized XGB classifier with SMOTEENN achieves a peak accuracy of 99.811%. This top-tier performance surpasses several existing benchmarks, validating the effectiveness of integrating IG-PCA with advanced resampling and optimization strategies. This work contributes a lightweight, scalable, and highly accurate IDS, offering a practical and efficient solution for enhancing security in resource-constrained, next-generation medical IoT systems.