FORECASTING THE PARTICULATE MATTER 2.5: A CASE STUDY IN CHALOEM PHRA KIAT DISTRICT, SARABURI PROVINCE, THAILAND

Abstract

The airborne particulate matter, which has been especially concerned about the case of PM2.5, has been a prominent factor heavily affecting everyday lives of Chaloem Phra Kiat District's residents. It has negatively influenced the health of the residents, especially as it has become evident in allergic-related skin conditions among the residents. Having understood the decline in the quality of life because of PM2.5 pollution, this study has sought to make predictions for the amounts of PM2.5 concentrations based on the previous data over the last 4 months, 6 months, and 1 year. The forecasting models used in the analysis revolved around the Auto-Regressive Integrated Moving Average (ARIMA), Vector Auto Regression (VAR), and Long Short-Term Memory (LSTM). The Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) were used to estimate the accuracy and performance of these models employing MATLAB and Orange as the data analysis software. It was found that the error rates between the two models of ARIMA and VAR were comparable, while the LSTM model showed significantly lower error rates, with the lowest MAE of 1.67 µg/m³ and MAPE of 7.94%, which was indicative of a better capacity to forecast. Additionally, the study showed clear seasonal fluctuations in the PM2.5 concentrations, which grew steadily to peak during the winter, then fell in summer, and finally fell to their lowest during the rainy season. For example, the peak monthly average in January reached over 55 µg/m³, while in August, it dropped below 15 µg/m³. A consistent cyclical pattern was found every year. As a benchmark forecasting and comparative analysis, this research laid a foundation for further research studies, possibly using advanced machine learning algorithms for further improvement of predictive accuracy and robustness of the models involved.