Hybrid Ensemble – Deep Learning Framework for Chronic Kidney Disease Prediction

Abstract

Early detection of chronic kidney disease (CKD) is crucial, as delayed diagnosis can lead to irreversible damage and high mortality rates. Traditional diagnostic methods often struggle to capture the complex, non-linear interactions among diverse clinical indicators such as blood pressure, haemoglobin, blood glucose, and serum creatinine. To address this limitation, the present study evaluates the performance of a proposed CatBoost-DeepNet hybrid framework against several well-established machine learning classifiers. The framework integrates the feature selection and interpretability strengths of CatBoost with the representational capacity of a deep neural network, enabling more accurate and reliable CKD prediction. Based on a real-world dataset of 400 patient records and 25 clinical features, we compared CatBoost-DeepNet to nine traditional and ensemble models, i.e., K-Nearest Neighbors (KNN), Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting, AdaBoost, XGBoost, LightGBM, and Extra Trees. Results indicate that CatBoost-DeepNet consistently outperformed baselines with a test accuracy of 99.17%, precision of 99.50%, recall of 99.00%, and an F1 score of 99.25%. Confusion matrix assessment also confirmed the diagnostic reliability of the model with zero false negative and just a single false positive. These findings suggest that CatBoost-DeepNet is a strong, generalizable, and clinically valuable platform for early prediction of CKD.