Enhanced early chronic kidney disease prediction using hybrid waterwheel plant algorithm for deep neural network optimization

Abstract

Abstract Chronic Kidney Disease (CKD) is a progressive condition primarily caused by diabetes and hypertension, affecting millions worldwide. Early diagnosis remains a clinical challenge since traditional approaches, such as Glomerular Filtration Rate (GFR) estimation and kidney damage indicators, often fail to detect CKD in its initial stages. This study aims to enhance early CKD prediction by developing a deep neural network optimized with a novel hybrid metaheuristic that combines the Waterwheel Plant Algorithm (WWPA) with Grey Wolf Optimization (GWO). Using the UCI CKD dataset, rigorous preprocessing techniques-including data imputation, normalization, and synthetic oversampling-were employed to enhance data quality and mitigate class imbalance. A multilayer perceptron (MLP) regression model was trained and optimized through the WWPA-GWO framework and benchmarked against other optimization algorithms, including PSO, GA, and WOA. Results demonstrated that the standard MLP achieved moderate performance (MSE = 0.00177, RMSE = 0.0420, MAE = 0.0100, $$R^2$$ = 0.8793), whereas the optimized model achieved significant improvements (MSE = $$3.06 \times 10^{-6}$$ , RMSE = 0.00175, $$R^2$$ = 0.9730) with reduced computational time (0.0999 s). Statistical validation using ANOVA ( $$p < 0.0001$$ ) and Wilcoxon signed-rank testing ( $$p = 0.002$$ ) confirmed the robustness of the approach. These findings highlight the effectiveness of the WWPA-GWO hybrid optimization strategy for deep neural networks, offering a reliable and efficient pathway for early CKD detection. Future work will explore the integration of advanced imputation methods, multi-modal data sources, and federated learning frameworks to enhance the model’s generalizability and clinical utility in diverse healthcare settings.