Machine learning for early detection of chronic kidney disease
Chronic kidney disease (CKD) affects over 850 million people globally and is strongly associated with hypertension, diabetes, and genetic predisposition. Its asymptomatic early stages complicate diagnosis, while reliance on glomerular filtration rate (GFR) and serum creatinine is limited by variability and workforce shortages. Machine learning (ML) offers scalable diagnostic solutions, though many studies emphasize accuracy over feature selection, interpretability, and theoretical grounding. Using a 400-patient dataset from the UCI CKD repository, missing values were imputed with feature means and data normalized via Min-Max scaling. A two-step feature selection strategy was employed, combining Pearson correlation (≥0.5 with the target) and model-based feature importance derived from ensemble methods such as Random Forest, ensuring both statistical robustness and clinical interpretability. Nine key predictors, including hemoglobin, packed cell volume, serum creatinine, sugar, hypertension, and diabetes mellitus, were retained. Seven classifiers, Multilayer Perceptron (MLP), Random Forest (RF), Logistic Regression (LR), Decision Tree (DT), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Gaussian Naïve Bayes (GNB) were evaluated using 5-fold cross-validation, with accuracy, confusion matrices, and AUROC as metrics. RF achieved the highest accuracy (99%), while GNB attained the strongest AUROC (0.993). All models performed robustly (94-99% accuracy; AUC ≥ 0.580). Feature selection improved efficiency, reduced redundancy, and highlighted clinically relevant predictors. This study advances CKD prediction by integrating feature engineering, supervised learning, and nephrological theory. Findings highlight ML’s potential for early, cost-effective detection, complementing traditional diagnostics. Future work should expand to larger, diverse datasets and pursue external validation to ensure generalizability and seamless clinical integration.