Replication Materials for: "Explainable Machine Learning for SDG 3: Nonlinear Patterns, Domain Specificity, and Information Structures Across OECD Health Outcomes"
Abstract
Description: This repository contains the complete empirical replication pipeline for the research paper titled “Explainable Machine Learning for SDG 3: Nonlinear Patterns, Domain Specificity, and Information Structures Across OECD Health Outcomes”. The study investigates the predictive performance and explainable drivers of health outcomes across 38 OECD countries (2000–2023), leveraging multi-source longitudinal data from the World Bank (WDI), the World Health Organization (WHO), and the Worldwide Governance Indicators (WGI). Methodological Highlights: Predictive Framework: Implementation of comparative linear baselines (OLS/Ridge) alongside non-linear machine learning ensembles (Random Forest and XGBoost). Temporal Validation: A rigorous rolling expanding-window temporal validation protocol (2021–2023) to assess model stability. Explainable AI (XAI): Utilization of TreeSHAP for global and local feature-importance attribution, including cross-outcome rank-correlation analysis. Robustness: Analytical verification using country-level cluster bootstrapping and multiple random seeds (42, 101, 2024, 77, 999). Repository Contents: Explainable Machine Learning for SDG 3.ipynb: The complete Python notebook containing the full analytical pipeline, from data integration to final XAI visualizations. SDG3_OECD38_MASTER_ANALYTICAL_UNIVERSE.csv (Primary Input): The master analytical dataset. Note: Please ensure this file is placed in the same directory as the notebook to allow proper execution. SDG3_OECD38_STAGE10_METADATA.json & SDG3_OECD38_STAGE10_INTEGRATION_AUDIT.xlsx: Supporting files documenting the data harmonization process and audit trails. Requirements: The code is written in Python (3.9+). Necessary dependencies include: pandas, numpy, scikit-learn, xgboost, shap, scipy, statsmodels, matplotlib, and seaborn. Citation: Please cite this dataset and the associated manuscript if used in your research.
