Title: Prediction of Heart Failure Patient Survival Using Artificial Neural Networks in JustNN
Authors: Husam Eleyan, Samy S. Abu-Naser
Volume: 10
Issue: 7
Pages: 19-27
Publication Date: 2026/07/28
Abstract:
Heart failure is a serious cardiovascular condition in which the heart is unable to pump enough blood to meet the body's needs. Predicting patient survival from clinical records can help identify high-risk cases and support clinical decision-making. This project applies an Artificial Neural Network (ANN) using the JustNN environment to predict the death event of heart failure patients during the follow-up period. The study uses the Heart Failure Clinical Records Dataset, which contains 299 patient records and 13 variables. Twelve variables are used as input features, including age, anaemia, diabetes, high blood pressure, creatinine phosphokinase, ejection fraction, platelets, serum creatinine, serum sodium, sex, smoking, and follow-up time. The target variable is DEATH_EVENT, where 0 represents survival during follow-up and 1 represents death during follow-up. The dataset was checked for missing values, binary variables were kept as 0/1, and continuous variables were normalized using Min-Max normalization before being saved as a CSV file and imported into JustNN. The final JustNN experiment used 211 training examples and 88 validation examples. The implemented network structure was 12-2-3-2-1, with 12 input nodes, three hidden layers containing 2, 3, and 2 nodes, and one output node. The final JustNN output reported 57,245 learning cycles, an average training error of 0.104680, and a validation result of 87.50% OK, corresponding to 77 correctly validated cases out of 88. The JustNN feature-importance output ranked follow-up time first, followed by ejection_fraction and serum_creatinine. These results are presented only as the outcome of the current JustNN experiment, and the inclusion of follow-up time is discussed carefully during interpretation.