Prediction of PM2.5 and PM10 in Chiang Mai Province: A Comparison of Machine Learning Models

Abstract

This study aimed to investigate the seasonal patterns and trends of particulate matter (PM2.5 and PM10) concentrations and to compare the prediction performance of three machine learning models for particulate matter trends in Chiang Mai province, Thailand. The trends and seasonal patterns were analyzed using the cubic spline function of PM2.5 and PM10 from 2012 to 2021. The prediction performance of three models, which included support vector machine (SVM), artificial neural network (ANN), and multiple linear regression (MLR), was compared by looking at the root mean square error (RMSE), the mean absolute error (MAE), and mean absolute percent errors (MAPE). The models with the lowest RMSE, MAE, and MAPE are considered the most suitable. The trend of PM2.5 and PM10 concentrations in Chiang Mai province had been determined to be high in March. PM2.5 and PM10 concentrations in Chiang Mai province have been slightly decreasing during the last 12 years. A model comparison revealed that SVM was the best model for predicting PM2.5, while MLR was the best model for predicting PM10.