One year mortality prediction in heart failure using feature selection and missing value imputation in deep learning

Heart failure is a disease caused by a deterioration in the function of the heart and a failure to supply the blood properly needed for the body. Follow-up measure with drugs and hospitalization can affect the survival of heart failure patients. Currently, none of the heart failure survival models including theses variables are effective yet. In this paper, we propose a method to effectively predict deaths within a year in patients with heart failure in Korea through preprocessing and deep learning. We used Korea Acute Myocardial Infarction Registry dataset which considers various features of patients with a left ventricular ejection rate 40% or less. Feature importance was measured using four models to find key features related to patients’ survival. We conducted several data preprocessing such as missing value imputation. Our machine learning approach showed higher accuracy than existing methods for predicting one year mortality of patients in heart failure.