Presentation Details
Explainable Machine Learning-Based Prognostic Modeling of Cardiac Death in Ischemic Heart Disease Using SHAP
ID：44
View Protection：ATTENDEE
Updated Time：2026-07-28 08:08:06
Hits：24
Online
Start Time：2026-07-31 11:55（Asia/Kolkata）
Duration：15min
Session：S6 Artificial Intelligence Use Cases »
Abstract
Ischemic heart disease is one of the leading causes of cardiac death worldwide. So accurate prediction and a timely prognostic model for clinical intervention is essential now. Though machine learning models predict risk effectively by utilizing multiple features, most of those rely on static prediction which fails to capture time-to-event information that is evolving patient risk over time. In addition, ML models often struggle to capture non-linear relationships and temporal dependencies present in survival data, which makes them less effective in clinical applications. Our study proposes Random survival forest (RSF) along with the Cox Proportional Hazards (Cox PH) model to overcome these limitations, and to improve prediction accuracy and interpretability. Unlike other models, our proposed model uses survival-specific features that provide time-to-event information, which results in a more dynamic understanding of patient risk progress. SMOTE and bootstrap-based resampling techniques are used to handle class imbalance and to improve model robustness, which was evaluated within the RSF framework. A publicly available dataset, which consists of 3987 IHD patients is used to evaluate our model. It uses 18 clinical variables, along with survival features such as event occurrence and follow-up time. Survival-specific evaluation metrics such as the concordance index (C-index) are used for the evaluation of our model’s performance. Our model achieves a C-index of 0.811. Our model is also compared with the baseline Cox PH model to check whether it gives better predictions. Our experiment results show that the proposed model outperforms the statistical model in capturing complex non-linear relationships and it provides more accurate and reliable risk stratification. The proposed model achieves an AUROC of 0.816, a precision of 0.631, and a recall of 0.744. The integration of handling imbalance data and a survival-aware model has significantly improved the prognostic accuracy. Overall, this study offers a robust model for time-to-event prediction for supporting clinical decisions and advancing precision cardiology through modern survival analysis techniques.— Ischemic heart disease is one of the leading causes of cardiac death worldwide. So accurate prediction and a timely prognostic model for clinical intervention is essential now. Though machine learning models predict risk effectively by utilizing multiple features, most of those rely on static prediction which fails to capture time-to-event information that is evolving patient risk over time. In addition, ML models often struggle to capture non-linear relationships and temporal dependencies present in survival data, which makes them less effective in clinical applications. Our study proposes Random survival forest (RSF) along with the Cox Proportional Hazards (Cox PH) model to overcome these limitations, and to improve prediction accuracy and interpretability. Unlike other models, our proposed model uses survival-specific features that provide time-to-event information, which results in a more dynamic understanding of patient risk progress. SMOTE and bootstrap-based resampling techniques are used to handle class imbalance and to improve model robustness, which was evaluated within the RSF framework. A publicly available dataset, which consists of 3987 IHD patients is used to evaluate our model. It uses 18 clinical variables, along with survival features such as event occurrence and follow-up time. Survival-specific evaluation metrics such as the concordance index (C-index) are used for the evaluation of our model’s performance. Our model achieves a C-index of 0.811. Our model is also compared with the baseline Cox PH model to check whether it gives better predictions. Our experiment results show that the proposed model outperforms the statistical model in capturing complex non-linear relationships and it provides more accurate and reliable risk stratification. The proposed model achieves an AUROC of 0.816, a precision of 0.631, and a recall of 0.744. The integration of handling imbalance data and a survival-aware model has significantly improved the prognostic accuracy. Overall, this study offers a robust model for time-to-event prediction for supporting clinical decisions and advancing precision cardiology through modern survival analysis techniques.
Keywords
Ischemic Heart Disease, Cardiac Death Prediction, Random Survival Forest, Cox Proportional Hazards, Survival Analysis, Time-to-Event Modeling.
Speaker
Priyadharshini S
Student Sastra UniversitySubmission Author
All Comments
Important Date
-
Conference Date
Jul 30
2026
toAug 01
2026
-
Jul 28 2026
Registration deadline
-
Jul 30 2026
Draft paper submission deadline
Sponsored By
The United Societies of Science
Organized By
Kongunadu College of Engineering and Technology
Supported By
IEEE Section
IEEE Madras Section
IEEE Madras Section
Contact Information
Previous Conferences
-
2026-07-30 India Trichy
2026 International Conference on Networks Computers and Communications
Submit Comment