A COMPARATIVE REVIEW OF METHODS FOR FORECASTING AMBIENT AIR QUALITY INDICATORS USING ARTIFICIAL INTELLIGENCE

The paper compares 21 reviews on forecasting the air quality index (AQI) and pollutant concentrations with machine learning. A survey of 155 studies from 2011-2021 serves as the reference source and is complemented by open-access reviews from 2018-2026. The reviews and the primary studies they cover are summarised by country, database, publication year, model and metric. Three stages in the development of the methods are identified, the reasons for disagreements between reviews are explained, early studies from Uzbekistan are outlined, and recommendations on model selection for the region are given. The selection of the reviews is documented according to PRISMA 2020, and each review is appraised for risk of bias. A model selection criterion is also proposed that combines a skill score against the persistence forecast, a data-sufficiency index, the stability of feature attributions and computing cost, on hourly data from three stations in Tashkent it picked the best model family in 5 of 9 cases and gave no recommendation for the 24-hour horizon, where no model was reliable.