การทำนายคุณภาพอากาศในประเทศไทย: การเปรียบเทียบวิธี ARIMA และ Machine Learning สำหรับ PM2.5 และ PM10

Abstract

This study examined the seasonal patterns and trends of particulate matter (PM2.5 and PM10) concentrations in Thailand and compared the prediction performance of five Models, including Autoregressive Integrated Moving Average (ARIMA), Autoregressive Moving Average with Exogenous Variable (ARIMAX), a multiple linear regression (MLR), an artificial neural network (ANN), a support vector machines and a random forest (RF). Using the cubic spline function, trends and seasonal patterns from 2010 through 2021 were explored. ARIMA, ARIMAX, MLR, ANN, SVM and RF models were utilized to analyze PM2.5 and PM10, was compared by looking at the root mean square error (RMSE), the mean absolute error (MAE), mean absolute percent errors (MAPE), and R-Squared (R2 ). The highest levels of PM2.5 and PM10 from 2010 through 2021 in the North were recorded between February and April, in the Northeast had been determined to be high in January to March, the highest levels in the Central from November to March, and no seasonal patterns of PM2.5 and PM10 were observed in the Southern of Thailand. The results of PM2.5 and PM10 prediction accuracy performance between ARIMA and the machine learning methods considers the day, year, temperature, relative humidity, barometric pressure and wind speed factors by split data of training and testing into 50:50, 60:40 and 70:30 the results demonstrated that the ARIMAX model outperformed the other models in predicting PM2.5 and PM10 levels that all 4 regions in Thailand.