Trustworthy and Ethical AI for Intrusion Detection in Healthcare IoT (IoMT) Systems: An Agentic Decision Loop Framework

Trustworthy and Ethical AI for Intrusion Detection in Healthcare IoT (IoMT) Systems: An Agentic Decision Loop Framework📋 Overview The study presents a multi-agent intrusion detection architecture that integrates:A supervised flow-based detectorA Deep Q-Network (DQN) triage agentA NIST AI RMF–aligned ethical rule engineThe framework enables trustworthy, safe, and context-aware intrusion detection in healthcare IoT environments (IoMT).🏗️ Repository Structure agentic-ethical-ids-healthcare/ │ ├── src/ # Source code for model, rule engine, and agent │ ├── train_agent.py │ ├── ethical_engine.py │ ├── detector_model.py │ └── utils/ │ ├── data/ # Links or sample data subsets │ ├── CIC-IoMT-2024/ │ └── CSE-CIC-IDS2018/ │ └── MIMIC-IV/ │ ├── notebooks/ # Jupyter notebooks for training and analysis │ ├── models/ # Pretrained model checkpoints (.pth, .pkl) │ ├── results/ # Evaluation outputs and figures │ ├── requirements.txt # Python dependencies ├── LICENSE # MIT License for open research use └── README.md # Project documentation ⚙️ Setup and InstallationClone the repository and set up your environment: git clone https://github.com/ibrahimadabara01/agentic-ethical-ids-healthcare.git cd agentic-ethical-ids-healthcare python -m venv venv source venv/bin/activate # On Windows: venv\Scripts\activate pip install -r requirements.txt OR You can download the winrar version so you get all files in one folder 📊 DatasetsThis project uses three datasets:DatasetPurposeSourceCIC-IoMT 2024Primary IoMT intrusion detection datasetCanadian Institute for CybersecurityCSE-CIC-IDS2018Domain-shift evaluationCIC Dataset PortalMIMIC-IV (Demo)Clinical context signalsPhysioNet⚠️ Note: All datasets are publicly available. The MIMIC-IV Demo contains only de-identified data. should incase you have any challanges downloading the datasets i will be availabe to help you through the process linkedin: https://www.linkedin.com/in/ibrahimadabara/Email: ibrahimadabara02@gmail.com🚀 How to Reproduce ResultsRun the full pipeline (training + evaluation): python src/train_agent.py --config configs/agentic_ids.yaml This script:Trains the supervised flow-based detector on CIC-IoMT 2024Fine-tunes the DQN triage agentEvaluates under domain-shift using CSE-CIC-IDS2018Computes Ethical Compliance Rate (ECR), False Escalation Rate (FER), and CAS metricsThe MIMIC-IV📈 Key MetricsMetricDescriptionAccuracyCorrect classification rate across all flowsF1-Score (Weighted)Balanced measure of precision and recallEthical Compliance Rate (ECR)Percentage of actions consistent with governance rulesFalse Escalation Rate (FER)Proportion of overreactions (false alarms)Contextual Adaptation Score (CAS)Robustness under domain-shift📘 CitationIf you use this repository, please cite: Adabara, I. M., et al. (2025). Trustworthy and Ethical AI for Intrusion Detection in Healthcare IoT (IoMT) Systems: An Agentic Decision Loop Framework. 🔒 Ethical ComplianceAll experiments comply with PhysioNet and HIPAA de-identification standards. The MIMIC-IV Demo dataset was used under credentialed access and contains no PHI.🧾 LicenseThis project is released under the MIT License, allowing free use for research and educational purposes.