A Lightweight Hybrid Model for Early Chronic Kidney Disease Prediction in Resource-Constrained Healthcare

Early diagnosis of chronic kidney disease (CKD) remains challenging because the disease may progress with few symptoms until substantial renal impairment has occurred.This study presents an optimised hybrid framework that integrates principal component analysis (PCA) with an artificial neural network (ANN) for CKD prediction using the University of California, Irvine, CKD dataset.The dataset comprised 400 patient records with 24 clinical predictors and was divided using a stratified 80:20 split into a 320-record development set and an 80-record locked independent hold-out set.Model selection and internal validation were performed using three repetitions of five-fold nested cross-validation, with three-fold inner cross-validation for preprocessing, PCA, ANN, and decision-threshold tuning.Across the 15 outer validation folds, the optimised PCA-ANN achieved a mean accuracy of 97.60% ± 2.03% (95% CI: 96.48%-98.73%)and a balanced accuracy of 97.69%.On the locked hold-out set, the final model achieved 98.75% accuracy (95% CI: 93.25%-99.78%),98.00% recall, 98.99% F1-score, and a receiver operating characteristic-area under the curve of 1.000, with one false-negative CKD case.The proposed technique shows promising results for the early detection of CKD but needs to be validated on large-scale, more diverse clinical datasets before deployment in real-world settings Categories: Advances in Biomedical data analysis, AI/ML-based decision support systems, Machine Learning (ML)