Artificial Intelligence for Chronic Kidney Disease Risk Stratification in the USA: Ensemble vs. Deep Learning Methods

Abstract

This research provides the application of machine learning and deep learning techniques in early detection of chronic kidney disease (CKD) using clinical data which are commonly collected in U.S. healthcare settings. CKD, a progressive condition marked by declining kidney functions, poses a major public health challenges in the U.S due to CKD’s prevalence and the high cost of treatment in after stages. By utilizing a dataset comprising 24 clinical parameters from 400 individuals 250 of whom were diagnosed with chronic kidney disease the research emphasizes the critical need for early and accurate prediction to update patient outcomes and minimize the burden on the healthcare system. The methodology of the research included data preprocessing, imputation of missing values of the CKD, and strategic feature selection, which are followed by the implementation of various machine learning algorithms such as K-Nearest Neighbors and Gradient Boosting, beside it deep learning models including Convolutional Neural Networks (CNN) and Artificial Neural Networks (ANN). Among these, Gradient Boosting emerged as the most effective approach, achieving an impressive 97% accuracy in predicting CKD status in healthcare system in U.S. Its performance highlights the potential of machine learning in identifying key diagnostic features of CKD and also offering a suitable solution for early intervention in clinical practice across the whole U.S.