Datasets and Code for: Generalization and Robustness in CKD Risk Prediction Across Independent Datasets

This repository contains the datasets, code, and experimental materials used in the study “Generalization and Robustness in CKD Risk Prediction: A Comparative Study Across Independent Datasets.” The objective of the study is to investigate how machine learning models for chronic kidney disease (CKD) prediction generalize across independent clinical datasets collected in different healthcare environments. The repository includes the harmonized versions of two CKD datasets originating from Al-Ramadi Teaching Hospital (Iraq) and Apollo Hospitals, Managiri (India). To enable cross-dataset evaluation, the datasets were aligned through a feature harmonization process that produced a shared set of clinical variables used in all experiments. The repository also provides the Jupyter notebooks used to perform model training, within-dataset evaluation, cross-dataset testing, calibration analysis, and SHAP-based interpretability experiments. The notebooks implement five machine learning models commonly used for tabular clinical data: Decision Tree, Random Forest, XGBoost, Support Vector Machine, and Multilayer Perceptron. In addition to the code and datasets, the repository includes the experimental outputs and figures generated during the study, including performance tables, calibration analyses, generalization gap measurements, and SHAP feature-importance visualizations. All materials are provided to support transparency, reproducibility, and reuse of the experimental framework for evaluating machine learning models under dataset shift in clinical prediction tasks.