Tools
Tools
Joseph, Alan (2025) PulseGuard: A Two-Stage Explainable Machine Learning Framework for Heart Disease Detection and Severity Prediction. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (976kB) | Preview |
Preview |
PDF (Configuration Manual)
Download (4MB) | Preview |
Abstract
Heart diseases are also a major cause of death in many countries of the world, and hence there is necessity to detect and preemptively gauge the risk. The study describes PulseGuard, a two-step machine learning system that enhances the accuracy, interpretability, and clinical utility of heart disease predictions. The system combines two publicly available datasets namely, the Kaggle Cardiovascular Dataset to perform binary detection and the UCI Heart Disease Dataset to perform multi classes severity grading. Following the extensive preprocessing, cleaning of data, normalization, encoding and balancing with SMOTE, several models were tested at both phases.
In Stage-1 (Disease Detection), ensemble and neural models showed good results, with XGBoost having 73.3% accuracy, which was better than MLP and SVM. This step is dependable in separating diseases and non-diseases and as the basis of the diagnostic process. Stage-2 (Severity Classification), the task complexity is multi-class and the data is limited which led to further difficulties but the XGBoost was the most effective with an accuracy of 58.7 and the highest macro-F1 rate of all the models. In order to maximize clinical transparency, SHAP based interpretability was utilized, which demonstrated the existence of important predictors such as age, systolic blood pressure, cholesterol and glucose levels. The trained models were eventually deployed in a web-based clinical decision-support system, which allowed real-time prediction to be made by doctors. This report indicates that PulseGuard is an excellent and scalable means of supporting the early cardiovascular risk assessment that is interpretable.
| Item Type: | Thesis (Masters) |
|---|---|
| Supervisors: | Name Email Simiscuka, Anderson UNSPECIFIED |
| Subjects: | R Medicine > Diseases R Medicine > Healthcare Industry Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning |
| Divisions: | School of Computing > Master of Science in Data Analytics |
| Depositing User: | Ciara O'Brien |
| Date Deposited: | 07 Sep 2026 13:05 |
| Last Modified: | 07 Sep 2026 13:05 |
| URI: | https://norma.ncirl.ie/id/eprint/9865 |
Actions (login required)
 |
View Item |