Robust chronic kidney disease prediction via synthetic data validation and interpretable ensemble learning

Area of Science:

• Artificial Intelligence
• Machine Learning in Healthcare
• Data Science

Background:

• Early diagnosis of Chronic Kidney Disease (CKD) is crucial for effective patient management.
• Existing diagnostic methods may face challenges with data scarcity or interpretability.

Purpose of the Study:

• To develop an interpretable ensemble learning framework for accurate CKD prediction.
• To leverage synthetic data augmentation and explainable AI techniques to enhance model performance and clinical utility.

Main Methods:

• Integration of Conditional Tabular Generative Adversarial Network (CTGAN) for synthetic data generation.
• Application of Recursive Feature Elimination with Cross-Validation (RFECV) for feature selection.
• Utilization of SHapley Additive exPlanations (SHAP) for model interpretability.
• Evaluation of ensemble classifiers (Random Forest, XGBoost, LightGBM, CatBoost) on original and augmented datasets.

Main Results:

• Augmented models achieved up to 99.17% test accuracy, with significant improvements in precision, recall, and F1-score.
• Statistical, predictive, and visual evaluations confirmed the high quality of CTGAN-generated synthetic data.
• SHAP analysis provided clinically meaningful explanations for CKD predictions.

Conclusions:

• The proposed interpretable ensemble learning framework effectively enhances CKD prediction using augmented data.
• Explainable AI (SHAP) offers valuable clinical insights, supporting reliable and transparent diagnostic tools for Chronic Kidney Disease.