Prediction of Mortality in Liver Transplant Recipients Using Neural Network

Predicting mortality risk in liver transplantation is an essential procedure that can help inform medical professionals in decision-making, risk stratification, and patient selection. There are scoring systems that have been used in the past to predict mortality rate of patients, but they have some limitations. Recently, machine learning models have become useful for predicting mortality after liver transplantation. This study focused on the feed-forward neural networks (FNN) to predict the risk of death following liver transplantation. A correlation heatmap was employed to perform optimal feature interrelationships between features in the dataset and the recursive feature elimination (RFE) was used to select the most relevant features and reduce the data dimensionality, improving prediction accuracy. minmax Scale was used to standardize and scale the selected feature for robust result. The FNN model and other models were trained and validated. The performance of the FNN was compared with other state-of-the-art algorithms for a comprehensive evaluation, which includes logistic regression, decision trees, gradient boosting, and random forests. This study shows that the feed-forward neural network performed better than other models when using the RFE techniques, obtaining an impressive accuracy rate of approximately 95.5%, which was significantly higher than the accuracy rates of the other models.