Interpretable chronic kidney disease risk prediction from clinical data using machine learning

Chronic Kidney Disease (CKD) is a major cause of illness and death worldwide, with over 2 million cases diagnosed in the U.K. and potentially up to 1.8 million undiagnosed. However, there is a lack of longitudinal studies on CKD in India, resulting in limited data on its prevalence. CKD is often asymptomatic until 70% of the kidneys are severely damaged, and once this occurs, there is no cure. Patients may require dialysis or a kidney transplant to survive. Detecting the risk of CKD early is therefore crucial. In developing countries like India, many people cannot afford regular laboratory blood tests. This study aims to develop machine learning models to predict the likelihood of CKD using limited blood test results collected in India, including blood pressure, albumin, red and white blood cell count, blood urea, serum creatinine, HbA1Cs, and other biomarkers. Decision Trees and Logistic Regression classification algorithms were used, with hyperparameter tuning, achieving an F-score of 1. These promising results suggest that state-of-the-art results may be achievable with just six laboratory tests.

This version of the paper has been accepted for publication, after peer review and is subject to Springer Nature’s AM terms of use (https://www.springernature.com/gp/open-research/policies/accepted-ma...), but is not the Version of Record and does not reflect post-acceptance improvements, or any corrections. The Version of Record is available online at: http://dx.doi.org/10.1007/978-3-031-36402-0_63