PM2.5 Concentration Prediction Based on mRMR-XGBoost Model

Abstract

Air pollution is one of the main environmental pollution, in which air pollution component prediction is an important problem. At present, there have been many studies using machine learning methods to predict air pollution components. However, due to its numerous influencing factors and incomplete determination, there are still problems in accurate prediction. In this paper, the gas factors and meteorological factors collected by the self-developed integrated system are firstly used to construct the original feature set. Then, the mRMR algorithm is used to select data features from the perspective of maximum correlation and minimum redundancy. Finally, a prediction method of PM2.5 concentration in the next hour based on feature selection and XGBoost is designed by combining the data after dimension reduction with the XGBoost model. The experimental results show that mRMR algorithm can effectively select the features of air, and the prediction accuracy is improved even when only half of the features of the original data are used.

The work is supported in part by the State’s Key Project of Research and Development Plan under Grants (2019YFE0196400), Industry-University-Research Cooperation Project in Zhuhai (ZH22017001200072PWC), Guangdong Provincial Special Fund For Modern Agriculture Industry Technology Innovation Teams (No. 2021KJ122), and the Open Research Fund Program of Key Laboratory of Industrial Internet and Big Data, China National Light Industry, Bei**g Technology and Business University.