Heart disease remains the leading cause of death in the United States, with heart failure contributing significantly to these mortality rates. Defined by the CDC as the heart’s inability to pump sufficient blood and oxygen to support other organs, heart failure affects over 6 million Americans, with a five-year post-diagnosis mortality rate of nearly 50%. This study leverages deep learning to predict mortality risk in heart failure patients using a dataset of 299 records with 17 clinical features. Deep neural networks (DNNs) and convolutional neural networks (CNNs) were developed and optimized through hyperparameter tuning to enhance prediction accuracy. Model interpretability was achieved using SHapley Additive exPlanations (SHAP), which identified key predictors such as follow-up duration, ejection fraction, serum creatinine, and diabetes. These findings align with prior research highlighting ejection fraction and serum creatinine as critical factors, while also emphasizing time, diabetes, and age as significant predictors. This work demonstrates the potential of combining explainable AI with deep learning to support clinical decision-making in heart failure management.

This project was completed during a 10-week REU program at North Carolina A&T State University in Summer 2024. A huge thanks to my project partner, Jason Yin, our faculty advisor Dr. Letu Qingge, graduate student advisor Mr. Richard Annan, and Maxwell Sam for the work he’s done on this topic.