CGAN-Augmented XGBoost Framework for Early Detection of Chronic Kidney Disease: A Synthetic Data-Driven Approach

Abstract

Chronic kidney disease (CKD) affects about 10% of the world's population and is one of the main causes of death before age 65. It is very important to find the disease early and correctly so that it doesn't get worse and cause end-stage renal failure. Nonetheless, clinical datasets for CKD classification are inherently constrained in size, demonstrate class imbalance, and contain substantial missing values, which severely restrict traditional machine learning classifiers. This paper proposes a novel two-stage framework that integrates an Extreme Gradient Boosting (XGBoost) classifier with a Conditional Generative Adversarial Network (CGAN). In Stage 1, the CGAN makes high-quality patient records that statistically match real clinical distributions, fixing the class imbalance. XGBoost is trained on the augmented dataset in Stage 2. The proposed CGAN-XGBoost framework achieved a classification accuracy of 99.17%, a ROC-AUC of 0.994, a sensitivity of 97.73%, a specificity of 96.05%, an F1-score of 99.21%, and a Matthews Correlation Coefficient (MCC) of 0.983 when evaluated on the UCI CKD benchmark dataset. Comprehensive ablation studies and DeLong statistical significance tests (p < 0.001) confirm superiority over seven competing baselines.