Machine Learning Modeling of Vitality Characteristics in Historical Preservation Zones with Multi-Source Data

Abstract

Research on historic preservation zones (HPZs) has recently attracted increasing attention from academia and industry. With eight Beijing typical HPZs selected, this study evaluates critical vitality characteristics and identifies the key influencing factors via multi-source data and machine learning technology. The vitality characteristics were identified from three dimensions: physical space vitality, cyberspace vitality, and sentiment degree. For influencing factors, 23 variables were constructed from four aspects (morphological, functional, visual, and traffic) using Computer Vision (CV), natural language processing (NLP) and Geographic Information System (GIS) techniques. Then, three vitality dimensions were introduced as responsive variables to establish three Random Forest Regression models. Lastly, each factor’s influence degree and direction on vitality were explained based on the feature importance and correlation analysis. Through this study, we have thoroughly examined the different influencing factors of vitality in HPZs and summarized the following academic findings: (1) Density of road intersections, the number of shops, and road impedance are the three of the most significant influencing factors that are negatively related to vitality. (2) Factors that have the highest impact on the sentiment degree are road impedance and the number of public infrastructures, which also negatively affect the population’s satisfaction. (3) The number of catering and entertainment amenities are critical factors that positively affect cyberspace’s vitality. In this study, all three models have adequately explained variables and generalization capability, which can be applied to other larger HPZs in Beijing. In addition, the findings of this study can also potentially provide insights for enhancing precinct vitality and the governance of HPZs in other cities.