Predicting Life Expectancy in Upper‐Middle‐Income Countries Using Machine Learning: A Comparative Analysis

Creative Commons License

Süzen E., Kayakuş M.

SUSTAINABLE DEVELOPMENT, ss.1-17, 2026 (SSCI, Scopus)

  • Yayın Türü: Makale / Tam Makale
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1002/sd.71468
  • Dergi Adı: SUSTAINABLE DEVELOPMENT
  • Derginin Tarandığı İndeksler: Natural Science Collection (ProQuest), Social Science Premium Collection (ProQuest), Business Source Ultimate (EBSCO), Scopus, Materials Science & Engineering Collection (ProQuest), Political Science Database (ProQuest), Sociology Source Ultimate (EBSCO), Technology Collection (ProQuest), Social Sciences Citation Index (SSCI), IBZ Online, ABI/INFORM, Environment Index, Geobase, Greenfile, Index Islamicus, Political Science Complete, Public Affairs Index, Political Science Abstract (IPSA)
  • Sayfa Sayıları: ss.1-17
  • Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
  • Akdeniz Üniversitesi Adresli: Evet

Özet

This study aims to estimate life expectancy in upper-middle-income countries using socio-economic and health-related variables obtained from the World Bank's World Development Indicators database. To this end, four machine learning algorithms—Support Vector Regression (SVR), Random Forest, Gradient Boosting, and XGBoost—were applied; the models were compared using the 10-fold cross-validation method, employing the R2, RMSE, and MAE metrics. The analyses revealed that life expectancy is strongly associated with per capita GDP, health expenditure, and the urbanization rate. In the model comparisons, the highest predictive performance was achieved by the SVR model. Despite the limited sample size, SVR successfully captured non-linear relationships thanks to its kernel-based structure, demonstrating both the lowest error values and the most stable residual distribution. While the Random Forest model also performed well, the Gradient Boosting and XGBoost models achieved relatively lower accuracy. The findings demonstrate that machine learning methods provide effective tools for modeling multidimensional health indicators and that the SVR and Random Forest algorithms constitute robust alternatives for such data structures. The study aims to contribute to data-driven decision-making processes in global health and development policies.