Development of a Prediction Model for the Assessment of Air Quality in the City of Astana, Kazakhstan

Aruzhan Rakhimberdina, Gulden Ormanova, Didar Yedilkhan

2025

Closed access Metadata (CSL-JSON)
DOI 10.1109/e-tems64751.2025.11239309
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Air pollution is currently one of the pressing environmental issues in urban centers globally. Astana, the capital of Kazakhstan, has become one of the most polluted cities due to rapid urbanization over the past two decades. In this study, the use of traditional Multiple Regression (MR) is compared with cutting-edge Machine Learning (ML) models (Random Forest, CatBoost, LightGBM, Artificial Neural Networks (ANNs)) to define the most suitable prediction approach for air quality assessment. The key finding is that moving averages are more reliable predictors than lagging variables, especially in capturing short-term pollution trends. Meteorological parameters such as Wind Speed (WS), Temperature (T), and Relative Humidity (RH) significantly improved forecasting accuracy. I n addition, the study included atmospheric stability time series data, including boundary layer depth, vertical mixing coefficient, friction velocity, and horizontal mixing coefficient available through READY. The average concentration of PM2.5for the study period from January to December 2024 was 15.06 µg/m3. The results show that ANNs effectively capture complex patterns than Random Forest but slightly worse than MR. Among the ensemble methods, CatBoost and LightGBM offer the best balance between accuracy and interpretability. Sudden spikes in pollution from vehicles and industrial emissions remain difficult t o p redict due to the lack of real-time pollution tracking in different geolocations in Astana. This study confirms t hat M R with moving averages is still the most interpretable approach for predicting air quality in Astana. Future research is encouraged to explore hybrid models combining real-time pollution data and satellite imagery for better predictions.