Heart failure prediction with ensemble ML

For undergrad research, I built an ensemble-based classifier to predict 10-year heart failure risk using 4,238 records from the Framingham Heart Study. The model combined AdaBoost, CatBoost, XGBoost, and related learners, reaching 85.2% accuracy and 87.5% recall.

The exploratory analysis was as important as the model: age, blood pressure, cholesterol, and related attributes all needed careful handling on a skewed clinical dataset. The work was published at IEEE CONIT 2021 (DOI: 10.1109/CONIT51480.2021.9498561).

Focus

• Ensemble ML on longitudinal heart study data
• EDA-driven feature understanding for clinical prediction
• Peer-reviewed publication at IEEE CONIT