Ensemble model based on time series and regression to predict daily air quality in Tehran

Air quality index (AQI) is an important indicator for assessing urban air quality and its impacts on public health. Accurate AQI forecasting is essential for pollution management, especially in megacities such as Tehran, where air quality is strongly influenced by both human activities and meteorological conditions. This study proposes a hybrid ensemble framework combining autoregressive integrated moving average (ARIMA), generalized autoregressive conditional heteroscedasticity in mean (GARCH-M), multiple linear regression (MLR), and principal component regression (PCR) to predict daily AQI. Historical AQI and meteorological data from 2012 to 2015 were used for model calibration, and predictions were validated using 2016 observations. The results show that the proposed second ensemble model (Model 4) achieved the highest prediction accuracy across all seasons, with coefficients of determination (R²) ranging from 0.942 to 0.969 and RMSE values between 3.861 and 12.783 in the validation dataset. Compared to the standalone ARIMA model, the hybrid model improved R² by up to 42% and reduced NMSE by up to 82%. In winter, when air pollution was most severe, the model achieved an R² of 0.962 and RMSE of 6.019, demonstrating strong robustness under critical conditions. The findings confirm that integrating time series models with meteorological variables significantly enhances AQI forecasting accuracy and provides a reliable framework for operational air quality management in urban environments.