A Deep Learning Framework for Chronic Kidney Disease stage classification

Abstract

Chronic Kidney Disease (CKD) has become more prevalent, leading to a gradual decline in kidney function and, ultimately, in renal failure. Timely detection of the CKD stage is essential for enhancing healthcare services and decreasing morbidity and mortality. Hence, this study proposes a Metaheuristic-Hybrid Metaheuritstic eXplainable Artificial Intelligence (MHMXAI) driven Feature Selection (FS) approach and Deep Learning (DL) models for CKD stage prediction. MHMXAI approach selects the features with the highest scores from the Metaheuristic algorithm-Eagle Search Strategy, Hybrid Metaheuristic algorithm-Great Salmon Run-Thermal Exchange Optimization and eXplainable AI (XAI) tools like Local Interpretable Model-agnostic Explanations (LIME) and Shapley Additive exPlanations (SHAP) for their effectiveness. To evaluate the proposed method, eight DL models — Feedforward Neural Network, Recurrent Neural Network, Deep Neural Network, Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Bidirectional LSTM, Gated Recurrent Unit (GRU) and Bidirectional GRU were trained on selected features using different FS methods, as well as complete dataset. The models were assessed using performance metrics such as accuracy, precision, recall, F1-Score, Loss, Validation Loss and computation time. The CNN model outperformed others, achieving an accuracy between 98%-99.5% for all FS methods. Statistical tests, including the Friedman and Nemenyi post-hoc test, identified the CNN model trained with MHMXAI-selected features as the most robust choice for CKD stage prediction. These findings demonstrate that the proposed MHMXAI method effectively integrates metaheuristic algorithms and XAI tools, improving CKD stage prediction accuracy and clinical interpretability.