MACHINE LEARNING-BASED WALKABILITY MODELING IN URBAN LIFE CIRCLE

Abstract

With China's fast urbanization, the study of the walkability of residents' life circles has become critical to improve people's quality of life. Traditional walkability calculations are based on Lawrence Frank's theory. However, the weighted calculation method cannot be adapted to ever-changing and complicated scenarios as the scope and topic of research transforming. This study investigated walkability at the community level by combining machine learning techniques with multi-source data. Feature indicators affecting walkability were estimated from multi-source data. Machine learning was used to refine the weighting calculation under the previous indicator framework. We compared the performance of 20 regression models from 6 different machine learning algorithms for estimating the walkability of 14578 communities in downtown Shanghai. It is concluded that the Bagged Tree Model (R2=0.86, RMSE=0.36862) achieves the best performance, which is used to revise the initial walkability index values. The workflow proposed in this paper allows for rapid application of expert empirical consensus to comprehensive urban design and detailed urban governance in the future.