Enhanced Chronic Kidney Disease Prediction using Boruta Algorithm

Abstract

Chronic Kidney Disease (CKD) is a serious and life-threatening medical condition 
that impacts millions of individuals worldwide. It arises when both kidneys become 
damaged, impairing their ability to effectively filter blood. The kidneys play a 
critical role in maintaining overall health by filtering waste products and excess 
fluids from the blood, which are then excreted through urine. When kidney function 
declines, harmful toxins and fluids accumulate in the body, leading to a range of 
complications, including cardiovascular disease, anemia, bone disorders, and 
ultimately, kidney failure if left untreated. Recent studies highlight the growing 
burden of CKD, particularly in Sub-Saharan Africa, where the prevalence of the 
disease is alarmingly high. For instance, West Africa has reported a CKD prevalence 
rate of 16%, the highest on the continent. This underscores the urgent need for 
effective diagnostic and predictive tools to address the rising incidence of CKD in 
resource-limited settings. In recent years, machine learning techniques have gained 
significant traction in the field of medical research, particularly for disease 
prediction and diagnosis. These advanced computational methods offer the potential 
to improve early detection, enhance diagnostic accuracy, and support clinical 
decision-making. This paper explores the application of machine learning algorithms 
to predict CKD, with a focus on improving prediction accuracy. Among the various 
techniques evaluated, the Boruta algorithm, a feature selection method, 
demonstrated promising results by achieving satisfactory prediction outcomes even 
with a limited number of features. This highlights its efficiency in identifying the 
most relevant predictors of CKD. Furthermore, the Random Forest algorithm 
emerged as a highly effective model for CKD prediction, delivering exceptional 
performance metrics. Specifically, it achieved an accuracy of 99%, a precision of 
100%, an F1 score of 98%, a Cohen Kappa score of 97%, and a classification error 
of just 0.0125%. These results underscore the robustness and reliability of Random 
Forest in accurately classifying CKD cases, making it a valuable tool for early 
diagnosis and intervention.