Enhancing AQI forecasting accuracy: integrating ARIMA, ANN, and regression techniques with the development of HM4AQI web application

Air pollution is a major concern for human health, so accurate air quality prediction is essential for effective pollution management. This study proposes two hybrid models, ARIMA-ANN-REG and ARIMA-ANN-QREG, for forecasting daily air quality index (AQI) in Hat Yai, Songkhla, Thailand, and evaluates their performance against traditional models such as ARIMA and ANN, as well as the simpler hybrid ARIMA-ANN. The research utilized daily AQI data from January 2, 2013, to June 30, 2023, with 80% of the 3,013 observations used for training and the remaining 603 for testing. Six evaluation metrics were employed: root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), coefficient of determination ($$r^2$$), mean fractional bias (MFB), and mean fractional error (MFE). Results indicated that the ARIMA(1,1,1) model had strong performance with low RMSE (7.708), MAE (5.578), and MAPE (19.006%), along with a solid $$r^2$$ of 65% and minimal bias (MFB = 0.017, MFE = 0.181). While the ANN model achieved a slightly higher $$r^2$$ (65.71%), it also exhibited higher MAPE (21.553%) and MFB (0.077), suggesting a tendency to overpredict. Among the hybrids, ARIMA-ANN-REG performed better than ARIMA-ANN by incorporating linear regression, enhancing accuracy, and reducing bias. Notably, the ARIMA-ANN-QREG model achieved the lowest MAPE (18.647%), MFB (0.001), and MFE (0.179), demonstrating superior accuracy and minimal bias. To support practical application, a user-friendly web application was also developed to facilitate fitting these models to users’ time series data, making advanced forecasting tools accessible for environmental monitoring and decision-making.