Beat-Risk
Machine learning clinical prediction system for heart failure risk assessment utilizing DVC and Dagshub for end-to-end data versioning and reproducible MLOps.
Overview
Predictive health analytics pipeline trained on clinical indicators to forecast cardiovascular mortality risk with calibrated probability scores and feature importance explainability.
Architecture & Design
Ensemble ML models (XGBoost, Random Forest, Logistic Regression) with SHAP value interpretability. Complete MLOps data and model lineage tracked via DVC on Dagshub.
Key Capabilities
- Calibrated risk scoring with 89% ROC-AUC on validation cohorts
- SHAP-based feature importance breakdown for medical explainability
- Reproducible data and model pipeline managed with DVC