Accurate and efficient prediction of atmospheric PM1, PM2.5, PM10, and O3 concentrations using a customized software package based on a machine-learning algorithm.

Particulate matter (PM) and ozone (O3) pollution have been attracting increasing attention recently due to their severe harm to human health. PM and O3 are secondary pollutants, and there remain significant challenges in accurately and efficiently predicting their concentrations in the atmosphere. In this study, one-year monitoring data of PM1, PM2.5, PM10, and O3 concentrations as well as meteorological parameters and concentrations of various precursors (i.e., nitrogen oxides, SO2, CO, alkanes, aldehydes, and ketones) are obtained at a monitoring site in central China's Hunan province. The eXtreme Gradient Boosting model is trained and tested to achieve efficient and accurate predictions of the concentrations of the four pollutants. The effects of different datasets, input features, and model parameters on the prediction accuracy are investigated. Principal component analysis is employed to further reduce the dimensions of features, increasing the prediction efficiency. Finally, all model training and prediction processes are incorporated in an executable application using the PyQt5 framework to build a user interface. The customized software supports user-defined modeling. The software can simultaneously predict PM1, PM2.5, PM10, and O3 concentrations, making the prediction process convenient.