Problem
A university team project asking one question: can we predict a country's air pollution (PM2.5 exposure) from how it generates its electricity?
What I built
- The data cleaning pipeline that merges national electricity data with PM2.5 exposure into one clean table, plus the team's first baseline Random Forest.
- The Europe regional model: 41 countries, trained on 2000–2018 and tested on 2019–2022, with time-respecting cross-validation and models chosen on validation error only. It reached R² 0.86 on the test years, about 1.5 µg/m³ off on average, and beat the "same as 2018" baseline (R² 0.79).
- An ablation showing the electricity mix adds almost nothing: a control with only year and country scored about the same (R² 0.87). The skill comes from knowing the country and the trend.
- An Australia 2030 scenario with the team's Oceania model: if fossil fuels fell to 25% of electricity, three models disagreed even on the direction (−1.2 to +0.6 µg/m³), and 25% is outside Australia's historical range, so the effect is small and not reliable.
Team
A team project. My parts were the data cleaning, the Europe model and the Australia 2030 scenario.