APPLICATION OF MACHINE LEARNING METHODS FOR FORECASTING AIR QUALITY IN THE CITY OF BISHKEK

Abstract

Air pollution with fine particulate matter (PM2.5) remains one of the most acute environmental problems in Central Asian cities. Bishkek, located in a basin and subject to temperature inversions, records extreme pollution levels in winter every year. This paper presents a comprehensive analysis of the dynamics of PM2.5 concentrations and the Air Quality Index (AQI) for 2019–2024 based on monitoring data from the U.S. Embassy in Bishkek. Machine learning methods—Random Forest, Gradient Boosting, Logistic Regression—as well as an ensemble stacking model that integrates their advantages are applied to forecast the AQI and classify pollution levels. The models are compared using key metrics (R², RMSE, MAE, Accuracy, Precision, Recall, F1-score). The results demonstrate that the stacking approach provides more robust and accurate forecasting than the baseline models. The study confirms the effectiveness of integrating machine learning algorithms into an environmental monitoring and air-quality forecasting system.