KidneyTool: Un aplicativo web para predecir factores de riesgo crónico del riñón.

Abstract

Abstract: Chronic Kidney Disease (CKD) represents a significant public health problem in Ecua dor and worldwide, characterized by late detection and a high economic and social burden. This study develops a predictive model based on machine learning techniques to estimate the risk of CKD using clinical, biochemical, and lifestyle variables obtained from patients of the Ecuadorian Social Security Institute (IESS) in Loja, Ecuador. The methodological process included data cleaning and normalization, followed by feature selection using ANOVA-F and a review of relevant scientific literature. Based on this, several supervised models were trained and evaluated. Among them, the XGBoost Regressor achieved the best performance, reaching a coefficient of determination close to R² = 0.92, outperforming linear and regularized regression models. Additionally, an ordinal regression model was implemented to represent the hierarchical progression of the disease across five stages: normal, mild, moderate, severe, and kidney failure (G1 G5). This approach helped mitigate class imbalance and better reflect the clinical progression of CKD. The final model was integrated into a web-based tool called KidneyTool, developed using React, Flask, and Docker, which provides real-time risk predictions for potential clinical support.