Prediction of PM2.5 Concentration with Aerosol Optical Depth and Meteorological Factors in Southern Thailand Using Machine Learning Techniques

Abstract

Elevated PM2.5 levels are a critical indicator of poor air quality, with profound implications for public health and overall quality of life. The objectives of this study were (i) to investigate factors influencing the concentration of PM2.5 in southern Thailand and (ii) to compare the performance among multiple linear regression and machine learning techniques. This study examined twelve variables�CO, NO?, SO2, O?, PM10, wind speed, wind direction, temperature, relative humidity, air pressure, rainfall, and aerosol optical depth (AOD)�as potential factors influencing PM2.5 levels. Daily data from monitoring stations in two stations, Phuket station and Hat Yai City, Songkhla station, were analyzed for the period 2019�2022. Predictive models were conducted using multiple linear regression (MLR) and two machine learning techniques: artificial neural networks (ANN) and extreme gradient boosting (XGBoost). Additionally, a multiple linear regression (MLR) model was utilized as a fundamental tool to analyze the factors influencing PM2.5 concentration at each monitoring station. Model performance was evaluated based on root mean square error (RMSE), mean absolute error (MAE), and the coefficient of determination (R�). Splitting the data into training and testing sets of 70:30, the analysis revealed that at the Phuket station, MLR model outperformed both the ANN and XGBoost models. However, at the Hat Yai City station, the results were inconclusive. In other words, no specific model emerged as the best performer. Specifically, at the Phuket station, MLR achieved an RMSE of 2.4738, an MAE of 1.999, and an R2 value of 67.77%. However, it is important to acknowledge that model performance may vary based on factors such as the completeness of the datasets, geographical location, and the time period under consideration.

????? PM2.5 ??????????????????????????????????????????????????????????? ???????????????????????????????????????????????????????????????????? ?????????????????????????????? 1) ???????????????????????????????????????? PM2.5 ??????????????????????????? ??? 2) ?????????????????????????????????????????????????????????????????????? (Multiple Linear Regression) ??????????????????????????????? (Machine Learning Techniques) ???????????????????????????????????????? 12 ?????? ?????? ???????????????? (CO) ????????????????? (NO2) ????????????????? (SO2) ????? (O3) PM10 ?????????? ???????? ???????? ???????????????? ??????????? ??????????? ??????????????????????????????????????????? (Aerosol Optical Depth: AOD) ????????????????????????????????? PM2.5 ???????????????????????????????????????????? 2 ???? ?????? ?????????????????? ??? ????????????????? ???????????? ???????? ?.?. 2562�2565 (2019�2022)??????????????????????????????????????????????????????????? (Multiple Linear Regression: MLR) ??????????????????????????????? 2 ?????? ?????? ??????????????????? (Artificial Neural Networks: ANN) ??? Extreme Gradient Boosting (XGBoost) ??? MLR ??????????????????????????????????????????????????????????????????????? PM2.5 ??????????????????? ???????????????????????????????????????????????????????????????????? ?????? ???????????????????????????????????? (root mean square Error: RMSE) ???????????????????????????????? (mean absolute error: MAE) ????????????????????????????? (R�) ???????????????????????????????????????????????????????????????????? 70:30 ??????????????????? ???????????????????????? ???????? MLR ????????????????????????????? ANN ??? XGBoost ???????? RMSE ??????? 2.4738 ??? MAE ??????? 1.999, ?????? R� ??????? 67.77% ???????????? ??????????????????????? ??????????????????????????????????????????????????????????????????????????????? ??????? ??????????????????????????????????????????????????????????? ? ???? ??????????????????????? ???????????????????? ???????????????????????????