Deep Learning Algorithms for the Prediction of Renal Failure: A Case Study from Tanzania's Muhimbili National Hospital

Abstract

Renal failure occurs when kidney function fails, and the nephron is the major engine for renal function. Patients fail to detect kidney disease in its early stages, leaving them with just two options: kidney transplantation or renal dialysis, both of which are excessively expensive, with dialysis costing roughly 27,440 USD per year and kidney transplants costing 45,000 USD. As a result, developing the framework for the decision support system will aid doctors in reaching the study's goal. This study proposes that Deep learning be used to predict renal failure before it proceeds to the chronic stage. The data set was given by Tanzania's Muhimbili National Hospital. The framework predictor of renal failure was determined using six machine learning techniques (Logistic Regression, Linear Discriminant Analysis, K-Neighbors Classifier, Decision Tree Classifier, Gaussian Naive Bayes, and Support Vector Machine). The best performance was reported to be a Gaussian Naive Bayes and Decision Tree classifier with 100% accuracy, followed by a Support Vector Machine and Logistic Regression with 98.6% accuracy. Cross-validation, confusion matrix, and receiver operating characteristics were also used to evaluate the framework. The precision measurements, recall measures, and f1-score scores serve as the foundation for the renal failure framework's 100% accuracy performance. We proposed that Gaussian Naive Bayes and Decision Tree classifiers be used to predict renal failure. © 2023 IEEE.