A Machine Learning Approach for Predicting Heart Failure Risk Based on Physical Examination and Blood Biomarkers
- Blagojević, Anđela1
- Geroski, Tijana1
- Dašić, Lazar2
- Pavić, Ognjen2
- Jakovljević, Djordje3
- Preveden, Andrej4
- Velicki, Lazar4, 5
- Milovančev, Aleksandra4, 5
- Rutten, Frans6
- Nelissen, Anne Pauline6
- Barlocco, Fausto7
- Fornaro, Alessandra7
- Jimenez-Blanco Bravo, Marta8, 9
- Maier, Lars10
- Tafelmeier, Maria10
- Filipović, Nenad1
- 1. Faculty of Engineering, University of Kragujevac
- 2. Institute for Information Technologies, University of Kragujevac
-
3.
Coventry University
- 4. University of Novi Sad Faculty of Medicine
-
5.
Institute for Cardiovascular Diseases of Vojvodina
-
6.
University Medical Center Utrecht
- 7. Hospital Universitario Ramon y Cajal
- 8. CIBER Cardiovascular Diseases
- 9. Careggi University Hospital
-
10.
University Hospital Regensburg
Description
Heart failure (HF) is a major global health burden, and early risk stratification based on routinely available clinical data can help guide timely intervention. This study presents a machine learning approach for predicting heart failure risk using physical examination findings and blood biomarkers. An XGBoost classifier was trained and evaluated, achieving an accuracy of 0.97, a macro-averaged F1-score of 0.89, and a cross-validated ROC-AUC of 0.964, correctly identifying 82% of low-risk and 98% of high-risk individuals. These results demonstrate the potential of ensemble machine learning models to support clinical decision-making in heart failure risk assessment. This work was presented at the 16th International Conference on Information Society and Technology (ICIST 2026), Kopaonik, Serbia, and was carried out within the STRATIFYHF project.
Files
full_paper_ICIST2026_final.pdf
Files
(297.5 kB)
| Name | Size | Download all |
|---|---|---|
|
md5:c0c9360f98b356a72ae5f02a03410bd4
|
297.5 kB | Preview Download |