PM2.5 Prediction Based on XGBoost

Haze pollution is a serious weather condition which occurs frequently in mainland China. As there has been an increasing worldwide research interest around topics in environment protection and human health, PM2.5 concentration is regarded as a vital index to reflect the air quality. The current PM2.5 prediction algorithm has the problems of single index and insufficient features. In this paper, based on XGBoost, a PM2.5 prediction model is proposed. Such model adopts the random forest algorithm for feature selection. Then, the optimal feature subset that affects PM2.5 concentration is selected. The experiment is conducted on real dataset of daily air quality and weather observations in Beijing from 2017 to 2019. The results show that the daily average concentration of PM2.5 can be accurately predicted with root mean square error at 8.63, and correlation at 95.41%.