An intelligent ensemble machine learning model for early detection of chronic kidney disease in aging populations

Résumé fourni par la source

Chronic Kidney Disease (CKD) remains a pressing global public health concern, accounting for approximately 1.7 million deaths annually and disproportionately affecting aging and underserved populations. The increasing burden of CKD, particularly in low-resource settings, underscores the urgent need for early, accurate, and scalable diagnostic tools. This study proposes a hybrid mathematical and artificial intelligence (AI) framework for the early prediction of CKD, with a focus on supporting healthcare strategies in aging and resource-limited communities. Utilizing clinical data from a case-control study conducted in District Buner, Khyber Pakhtunkhwa, Pakistan, the framework incorporates a structured modeling pipeline that involves data preprocessing (feature extraction, missing data imputation, and categorical encoding) and class balancing via the Synthetic Minority Over-Sampling Technique (SMOTE). The proposed system integrates multiple machine learning algorithms, including logistic regression, feedforward neural networks, decision trees, support vector machines, and random forests, within a novel ensemble learning strategy designed to enhance diagnostic precision. Model robustness was assessed using three distinct train–test scenarios: (90%, 10%), (75%, 25%), and (50%, 50%). Performance evaluation employed six metrics: accuracy, specificity, sensitivity, Youden index, Brier score, and F1 score, supported by comprehensive graphical and statistical analysis. The ensemble model consistently outperformed individual classifiers, achieving a mean accuracy of 97.71%, specificity of 97.19%, sensitivity of 99.84%, Youden index of 86.55, Brier score of 1.43%, and F1 score of 98.19%. Support vector machines and random forests ranked second and third, respectively, while decision trees exhibited the lowest performance. To the best of our knowledge, this is the first ensemble-based predictive framework for CKD developed using clinical data from Pakistan. The system holds strong potential for integration into real-world biomedical decision support systems, particularly in aging and underserved populations, thereby contributing to early detection, enhanced care delivery, and optimized resource utilization in the management of chronic diseases.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.