Performance of Box Jenkins and Artificial Neural Networks models on Air Pollution Index in Klang

This study focuses on air pollution levels in Klang Valley, a region frequently experiencing poor air quality due to industrial activity, haze, traffic and urbanization. The performance of Box Jenkins (ARIMA) and Artificial Neural Network (ANN) models was compared using data collected from Malaysia’s Department of Environment (DOE) over five years (2019-2023). The ARIMA model analyzed Air Pollution Index (API) as the variable, while the ANN model utilized API as the output variable and six pollutants sulphur dioxide (SO2), particulate matter (PM10), (PM2.5), ozone (O3), nitrogen dioxide (NO2) and carbon monoxide (CO) as input variables. Under Box Jenkins procedure, six ARIMA models were tested, with ARIMA (4,1,1) selected as the best based on AIC and BIC. As an ANN model, networks with varying hidden nodes were evaluated, and the model with five hidden nodes was identified as the best, achieving highest R² and lowest RMSE. Comparisons between the best ARIMA and ANN models showed that ANN outperformed ARIMA, with lower RMSE (5.5502) and MAE (4.1799). In conclusion, both ARIMA and ANN models effectively predicted API, however an ANN model offering slightly better performance based on the root mean square error.