A Hybrid CNN-SVM Model with Attention Mechanisms for Chronic Kidney Disease Prediction

Abstract

Chronic Kidney Disease (CKD) represents a significant global health challenge, requiring early and accurate diagnosis to prevent progression to end-stage renal failure. This study addresses the limitations of traditional diagnostic methods and the "black-box" nature of standard deep learning by developing an explainable, attention-driven hybrid Convolutional Neural Network- Support Vector Machine (CNN-SVM) framework. The methodology integrates 1D-Convolutional layers with a custom attention mechanism, which serves as a dual-purpose tool for both automated feature extraction and Explainable AI (XAI). By quantifying the importance of specific clinical variables, the model moves beyond simple prediction to offer transparent, justifiable diagnostic insights. The architecture is completed by a Support Vector Machine that replaces the final Softmax layer to optimize decision boundaries through structural risk minimization, while SMOTE-based class balancing ensures robustness against dataset imbalances. The research findings demonstrate state-of-the-art performance, with the hybrid model achieving a benchmark accuracy, precision, recall, and F1-score of 100% (1.0000) on the test set, significantly outperforming a standalone CNN baseline of 93.33%. A core contribution of this work is the visualization of feature importance, which identified key clinical biomarkers— such as hemoglobin, specific gravity, and serum creatinine—as primary drivers for prediction, thereby bridging the gap between machine learning outputs and clinical intuition. These findings are significant because they provide a robust, transparent, and low-cost diagnostic utility— deployed via a Flask web interface—that offers clinicians a reliable and explainable "second opinion" for early CKD detection, particularly in resource-limited settings.