DMP: IoMT attack prediction with ML models

The project generated a new dataset from a Healthcare Monitoring System (HMS) testbed without using external sources. The data included network traffic and biometric measurements stored in CSV format, comprising 16,317 samples (9,790 normal and 6,527 attack samples) across 44 features. It was intended to support research in real-time intrusion detection and patient monitoring within healthcare networks. The dataset was made publicly available with DOI: 10.82556/cgrm-ph43 and persistent identifier (PID): f2ff22aa-8dca-466b-ba4a-7cbb1981b282, generated under the TUWARD account.

The dataset followed FAIR principles:

Findable through rich metadata (following Dublin Core standards).

Accessible via an open-access trusted repository using standard protocols without embargo.

Interoperable using CSV format and linked to cybersecurity taxonomies.

Reusable through detailed documentation, codebooks, and a CC-BY license.

Software scripts for data visualization, network monitoring, and intrusion detection were also developed and made openly available via GitHub (https://github.com/12436894/IoMT_framework) and archived with Zenodo DOI: 10.5281/zenodo.15296798.

A hybrid machine learning model combining Random Forest and Gradient Boosting classifiers via a Voting Classifier was implemented to predict intrusion labels. The process included preprocessing (PCA feature selection and encoding), model training/testing, and evaluation through accuracy scores, classification reports, confusion matrices, ROC and Precision-Recall curves. All outputs, including datasets, plots, and evaluation reports, were saved automatically.

The data and software management adhered to GDPR and relevant data protection regulations. Long-term preservation, data security (encryption, backups), and ethical handling of personal data were ensured following institutional and Horizon Europe guidelines.