Optimized Machine Learning Models for Early Prediction of Kidney Failure

Abstract

Kidney failure’s poor prognosis, prevalence, and progression to end-stage renal disease pose serious health risks and healthcare burdens, yet early detection and treatment save lives and decrease the demand on the healthcare system. On average, a person may survive only 18 days without functioning kidneys. Kidney failure can be predicted using machine learning approaches, and the optimal algorithms for accurate diagnosis can be identified by using clinical data to identify relevant variables. Our research uses clinical data to help medical professionals anticipate kidney failure using machine learning techniques. Using variables such as creatinine, diabetes, and hypertension, we employ machine learning to predict kidney disease and select the most informative parameters for precise prediction. To predict kidney failure from multiple parameters, various algorithms were used to develop predictive models. The most effective model was identified among Random Forest, decision trees, support vector machines, K Nearest Neighbor, and logistic regression. The modified Random Forest model outperformed the default model with an F1-measure of 99% and an exceptional prediction accuracy of 99.75%, according to the results. In the end, this will help doctors save lives by enabling early diagnosis of kidney disease, allowing early treatment, and proper monitoring of patients’ health before dialysis is necessary.