Prediction Of Population Density In Key Areas Of Beijing

Since the beginning of last year,the people of the whole country have jointly fought against the COVID-19,and achieved world-renowned anti-epidemic results.However,the epidemic has not yet ended,it still has a strong infectious capacity,and the flow of population has increased the risk of the spread of the COVID-19.This paper uses data analysis and machine learning methods to study the population density in the key areas of Beijing,which can help relevant departments find the problem of crowd gathering early,and has a very positive significance for epidemic prevention and control.In this paper,firstly,the data set is visualized to show the relationship between each feature and population density index,and then the feature is constructed to add new features to enhance the prediction effect of the model.Three kinds of Boosting algorithms XGBoost,Light GBM and Cat Boost are used for prediction,and Bayesian Optimization method is used for parameter adjustment.This method greatly improves the speed of parameter adjustment,and the prediction ability of the model is also greatly improved.In addition,this paper introduces the SHAP value to explain the importance of features,which can measures the main factors that affect the population density in the area.In order to further improve the prediction ability of the model,this paper also uses Averaging and Stacking methods to establish six fusion models,and finally finds that the fusion model which obtains the optimal fusion weight by PSO algorithm achieves the best prediction effect.Compared with the best-performing single model XGBoost,the MAE of the fusion model is reduced by 1.9%,the RMSE is reduced by 4.8%.This paper innovatively proposes the use of PSO algorithm to obtain the best fusion weight of the model,and the use of this method improves the prediction performance of the model very well.In addition,the highlights of this paper include:the introduction of SHAP value to explain the importance of features,the use of Bayesian Optimization method to adjust parameters,and the use of equal weight fusion method,weighted average fusion method,Stacking algorithm and other fusion methods.The introduction of SHAP value enhances the interpretability of the black box model,while the use of Bayesian Optimization parameter and multiple model fusion methods improves the prediction effect of the model.