India has the worst pollution status amongst all the polluted countries in the world, wherein several cities face severe environmental problems due to the rampant increase in pollution. For research investigation two cities, Gaya and Tirupati were selected to estimate the air pollution SO₂ trends by three different modelling techniques: M5P and artificial neural network (ANN) and bagged artificial neural network (BANN). Both cities are cultural centres and important pilgrimage destinations in India. Data were collected with 340 observations for Gaya and 1477 observations for Tirupati between 2017 and 2020 from Central Pollution Control Board (CPCB) and used to generate the SO₂ models. During the study, results of three models show that BANN is the best model for Gaya with coefficient correlation (CC) 0.8563, mean absolute error (MAE) 1.7575, root mean square error (RMSE) 2.9023, Nash–Sutcliffe efficiency (NSE) 0.7243, Willmott index (WI) 0.9154 and normalized root mean square error (SI) 0.2457 having a testing stage with input combinations, particulate matter (PM_2.5), nitric oxide (NO), NO₂ (nitrogen dioxide), NOx, carbon monoxide (CO), ozone (O₃), relative humidity (RH), wind speed (WS), wind direction (WD), solar radiation (SR), vertical wind speed (VWS) and absolute temperature (AT), and for Tirupati, also BANN is the best model with CC 0.8634, MAE 1.9088, RMSE 2.5891, NSE 0.7421, WI 0.9233 and SI 0.3409 having a testing stage with input combinations, PM_2.5, NO, NO₂, NOx, CO, O₃, RH, WS, WD, SR, VWS, AT and SO₂. The most influential parameters are WD and NO₂ for Gaya and NO₂ and O₃ for Tirupati. Graphical results also confirm that BANN-based model is best performing for both cities Gaya and Tirupati.