Abstract
This study compares machine learning models with OLS linear regression for estimating life expectancy using OECD Health Statistics from 25 European countries, 2000–2022. OLS linear regression, Random Forest, and Multilayer Perceptron models were trained on 2000–2018 data and evaluated on 2019–2022 observations. Predictors included GDP per capita, health expenditure, physician density, alcohol consumption, and population aged 65+. Tuned Random Forest performed best among fitted macro-indicator models, but only modestly improved on OLS linear regression. A simple persistence benchmark was far more accurate (test R² 0.903 versus 0.525), indicating interpretive rather than predictive value. Prediction accuracy was weaker in lower-life-expectancy countries, underscoring country-level conditions.