Explainable Artificial Intelligence (XAI)-Supported Machine Learning Models for Cardiovascular Disease Risk Prediction: A Comprehensive Performance Analysis

Aktaş S.

8. Uluslararası Thales Fen, Mühendislik, Mimarlık ve Matematik Kongresi, Kahire, Mısır, Cairo, Mısır, 18 - 20 Temmuz 2026, ss.705-717, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Basıldığı Şehir: Cairo
  • Basıldığı Ülke: Mısır
  • Sayfa Sayıları: ss.705-717
  • Marmara Üniversitesi Adresli: Evet

Özet

Abstract

Although machine learning algorithms have demonstrated high predictive performance for the early diagnosis of cardiovascular diseases (CVDs), the lack of transparency in their decision-making processes (the "black-box" problem) limits their integration into clinical practice. This study aims to develop an integrated clinical decision support model that not only predicts CVD risk with high accuracy but also enhances the transparency of its predictions through Explainable Artificial Intelligence (XAI) approaches. Within the scope of the study, a dataset comprising 918 patients and 11 independent clinical features was utilized. Random Forest, XGBoost, and LightGBM algorithms were trained following hyperparameter optimization using GridSearchCV. Based on comprehensive performance evaluations, the tree-based ensemble model Random Forest outperformed the other models, achieving an accuracy of 87.50%, an F1-score of 89.20%, and an Area Under the Curve (AUC) value of 0.923. The statistical robustness and generalizability of the proposed model were further demonstrated through 5-fold cross-validation, which yielded a mean accuracy of 81.69%. To overcome the opaque nature of conventional machine learning algorithms, the game theory-based SHAP (SHapley Additive exPlanations) method was integrated into the system. SHAP analysis mathematically identified ST depression (ST_Depression) and chest pain type (ChestPainType) as the most influential variables affecting the model's risk predictions. The strong agreement between the SHAP findings and established principles of cardiology confirms that the model has not merely memorized statistical patterns but has successfully learned clinically meaningful pathological relationships. Consequently, the proposed XAI-supported architecture demonstrates that artificial intelligence technologies are not intended to replace physicians; rather, they serve as reliable tools capable of providing healthcare professionals with data-driven, unbiased, and fully transparent second opinions in clinical decision-making.

Keywords: Machine Learning, Explainable Artificial Intelligence, SHAP, Cardiovascular Diseases, Clinical Decision Support Systems.