Predicting Heart Failure

I evaluated whether machine learning can improve mortality prediction for actuarial risk assessment using a 368-patient heart failure dataset with 60 clinical, demographic, and cardiovascular predictors. The analysis compared Logistic Regression, SGD SVM, and Random Forest models against actuarial concerns like risk segmentation, underwriting precision, model validation, interpretability, and governance.

After cleaning the data, reducing multicollinearity with VIF, and narrowing the model to seven key predictors, the Random Forest model produced the strongest result with 98.2% accuracy and a 0.99 ROC AUC. The project frames machine learning as a complement to traditional actuarial methods rather than a replacement, with clear caution around small-sample reliability and real-world model risk.