Explainable Machine Learning for PM2.5 Prediction and Comparative Country-Level Risk Assessment in Europe and Central Asia

Fine particulate matter (PM2.5) is among the most harmful air pollutants. This study asks whether national development indicators (income, government AI readiness, sustainable-development progress and perceived pollution) add predictive information to meteorology for annual mean PM2.5 in ten cities of Europe and Central Asia (2020–2025; 59 city-years). Because the observations are repeated measurements from only ten cities, models were evaluated by leave-one-city-out cross-validation (prediction for a city not used in training) and by a temporal split (training on 2020–2023, testing on 2024–2025). Uncertainty was estimated with a city-cluster bootstrap. These designs were used in addition to the random five-fold design used previously. Indicators that embed air-pollution information (the SDG Index and the Numbeo Pollution Index), as well as redundant or topographic variables, were excluded from the main model. Random five-fold validation overstated transferability: for the same predictors, R2 fell from 0.76 to 0.64 when whole cities were held out, and the tree ensembles degraded most (R2 0.22–0.55). Under leave-one-city-out validation, meteorology alone yielded R2 = 0.44 (RMSE 5.9 µg/m3); adding log GDP per capita and AI readiness raised R2 to 0.63 (RMSE 4.8 µg/m3), although the 95% bootstrap intervals were wide and overlapping. The development indicators varied mainly between cities and did not improve year-ahead prediction beyond meteorology. Relative humidity was the only predictor with a consistent within-city association. All ten city means exceeded the WHO guideline of 5 µg/m3, by factors of 2.2–6.8. The results are exploratory and support, at most, a modest between-city contribution of development indicators.

Authors

Publication Details

Journal
Urban Science
Published
2026-11-19
DOI
https://doi.org/10.3390/urbansci10110676
Primary Topic
Air Quality and Health Impacts
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Explainable Machine Learning for PM2.5 Prediction and Comparative Country-Level Risk Assessment in Europe and Central Asia

Armen Yu. Ghazaryan, Anna S. Kirakosyan, Cédric Stéphane Tekouabou Koumetio, Mareta Arshakyan et al.
4 links
Urban Science
Air Quality and Health Impacts
article

Explainable Machine Learning for PM2.5 Prediction and Comparative Country-Level Risk Assessment in Europe and Central Asia

Armen Yu. Ghazaryan, Anna S. Kirakosyan, Cédric Stéphane Tekouabou Koumetio, Mareta Arshakyan, Kristina Khudaverdyan, Gohar Zaprosyan
article en

Abstract

Fine particulate matter (PM2.5) is among the most harmful air pollutants. This study asks whether national development indicators (income, government AI readiness, sustainable-development progress and perceived pollution) add predictive information to meteorology for annual mean PM2.5 in ten cities of Europe and Central Asia (2020–2025; 59 city-years). Because the observations are repeated measurements from only ten cities, models were evaluated by leave-one-city-out cross-validation (prediction for a city not used in training) and by a temporal split (training on 2020–2023, testing on 2024–2025). Uncertainty was estimated with a city-cluster bootstrap. These designs were used in addition to the random five-fold design used previously. Indicators that embed air-pollution information (the SDG Index and the Numbeo Pollution Index), as well as redundant or topographic variables, were excluded from the main model. Random five-fold validation overstated transferability: for the same predictors, R2 fell from 0.76 to 0.64 when whole cities were held out, and the tree ensembles degraded most (R2 0.22–0.55). Under leave-one-city-out validation, meteorology alone yielded R2 = 0.44 (RMSE 5.9 µg/m3); adding log GDP per capita and AI readiness raised R2 to 0.63 (RMSE 4.8 µg/m3), although the 95% bootstrap intervals were wide and overlapping. The development indicators varied mainly between cities and did not improve year-ahead prediction beyond meteorology. Relative humidity was the only predictor with a consistent within-city association. All ten city means exceeded the WHO guideline of 5 µg/m3, by factors of 2.2–6.8. The results are exploratory and support, at most, a modest between-city contribution of development indicators.

Urban ScienceVol. 10(11)
Air Quality and Health Impacts
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon