Application and Performance Comparison of Tree Model in PM2.5 Concentration Prediction

With the acceleration of urbanization and the increase of industrial emissions, air pollutants have posed an increasingly serious threat to human health and environmental safety. This study uses an air pollution dataset collected from Southeast Asian countries, which contains multiple pollutant concentration indicators. Three tree planting models were developed to perform regression prediction on PM2.5 concentration: Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM). The relationships among variables were explored by cleaning the original data set and combining visualization. In the model results, all three models achieved good predictive capabilities, but the RF performed the best. The findings demonstrate that the tree model performs well when the data scale is medium and the feature correlation is poor. This paper further analyzes the possible reasons for the performance differences of the model and points out the limitations of the current research, such as insufficient feature dimensions and inadequate parameter tuning of the model. Finally, this paper puts forward improvement suggestions, including introducing larger-scale data, adopting deep learning methods and combining with spatial visualization platforms, to enhance the accuracy and practical application value of air pollution prediction.