Tools
Tools
Yarlagadda, Anudeep (2025) A Super Learner-Based Framework for Securing Medical IoT Networks Enhanced Intrusion Detection. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (945kB) | Preview |
Preview |
PDF (Configuration Manual)
Download (1MB) | Preview |
Abstract
Intrusion Detection Systems (IDS) are essential for safeguarding Medical Internet of Things (MIoT) networks, which handle sensitive biometric and network data in real-time. Traditional machine learning models like Logistic Regression, SVM, and Naïve Bayes often lack adaptability, accuracy, and explainability when applied to highly imbalanced and complex medical datasets. These limitations include high false negative rates, poor generalization, and limited interpretability. This study introduces a Super Learner-based framework to overcome these challenges using the WUSTL-EHMS-2020 dataset, which contains both biometric and network flow metrics. The proposed Super Learner is a stacking ensemble that combines the strengths of LightGBM, AdaBoost, and Random Forest classifiers. These base models generate out-of-fold predictions through 10-fold cross-validation, which are then stacked and used to train a meta-learner—Gradient Boosting Classifier—ensuring unbiased, high-quality learning. The ensemble structure captures complex patterns and interactions missed by individual models. After preprocessing, balancing with SMOTE, and feature selection, the Super Learner achieved the highest accuracy of 100%, significantly outperforming baseline models. Additionally, Explainable AI techniques like LIME were integrated to interpret predictions and ensure trust and transparency in medical applications. This framework sets a new benchmark in secure, interpretable, and high-performing intrusion detection for medical IoT environments.
Actions (login required)
 |
View Item |