A two-level random forest model for predicting the population distributions of urban functional zones: A case study in Changsha, China

Understanding population density at a fine spatial scale is beneficial for urban management and planning. Existing machine learning methods have been widely used to predict the population using regular grids. However, regular grids defined as basic units lack semantic information, and error autocorrelation may be ignored when using machine learning methods. As a result, the prediction accuracy may be impacted. Therefore, a two-level random forest method is proposed in this research based on error compensation to predict the population distribution of urban functional zones. The first-level random forest model is used to model the census variable and the covariables, and the second-level model is further applied to address the error term. By using this two-level model, population distribution can be predicted in urban functional zones identified based on an area-weighted POI proportion. Experimental results from Changsha, China, show that the determination coefficients and root mean square error of the proposed method are 0.90 and 7,436, respectively. These results are better than those from comparative methods, which demonstrates the effectiveness of the proposed method. A population density of 203 single and 387 mixed functional zones is finally obtained, which can provide useful urban management and planning information.