A Bagging-GBDT ensemble learning model for city air pollutant concentration prediction

Abstract

The air pollution problem has become a serious environmental problem facing many cities in China in recent years. In this paper, we introduced the Gradient Boosting Decision Tree (GBDT) method into the base learner training process of Bagging framework, and proposed a prediction model Bagging-GBDT based on Bagging ensemble learning framework. Based on this prediction model, we selected Beijing city of China as an example and established a PM2.5 concentration prediction model to forecast the PM2.5 concentration for the next 48 hours at a given time point. To measure the validity of the model, we also trained support vector machine regression models and random forest model to calculate three statistical indicators (RMSE, MAE and R2) for the proposed models on the test set to compare models performance. The experimental results show that our Bagging-GBDT model can better reduce the prediction bias and variance, and the prediction effect is better than SVR and random forest models.