Artificial Neural Network Based Prediction of PM2.5 Mass Concentration

Economic growth and human activities affect the increase of particulate matter
$(\text{PM}_{2.5})$
concentration. In addition to the sources of emissions, the mass concentration of
$\text{PM}_{2.5}$
may be impacted by meteorological factors such as temperature, humidity levels, atmospheric pressure, precipitation, and the speed/direction of the wind. A previous study established a monitoring system for air quality at Tokong Nanas (GKU) and the Deli Building of Telkom University, located in Bandung, using microsensor technology. Various forecasting techniques were also employed to predict the mass concentration of
$\text{PM}_{2.5}$
, considering its meteorological factors. However, the study found that not all the meteorological parameters give significant results in
$\text{PM}_{2.5}$
forecasting, with an RMSE value of 27
$\mu\mathrm{g}/\mathrm{m}^{3}$
. Hence, this study optimized the forecasting system of
$\text{PM}_{2.5}$
mass concentration using the Artificial Neural Network Backpropagation method. Only a few meteorological parameters were taken into consider-ation in the forecasting system, which has a significant impact on the forecast quality, such as rainfall intensity, relative humidity, and wind speed. As a result, the Tokong Nanas (GKU) has the best network model for 4-9-12-9-1 architecture with a learning value of 0.2, whereas the Deli Building is 4-20-9-9-1 with a 0.3 value. The RMSE and MAPE performances generated by GKU and Deli best network models were 8
$\mu \mathrm{g}\mathrm{m}^{3}$
at 37% and 13
$\mu \mathrm{g}/\mathrm{m}^{3}$
at 15%, respectively. Additional investigation is required to scrutinize the conduct of contaminated atmosphere and to tackle the predicament of air purity in Bandung Metropolitan in forthcoming times.