Machine Learning Analysis Framework for Passenger Distribution
This paper proposes a generalized machine-learning framework for predicting and analyzing travel patterns.Firstly,based on the gravity model,radiation model,and PWO model,the interpretable key influencing factors are selected as input variables for the model,and a multi-model comparison machine learning analysis framework is constructed to improve the generalization ability of the model.Further-more,the predictive performance and generalization ability of the analytical framework is tested using 1km grid scale data and street town scale data from Ningbo,transportation census area data and district da-ta from the Tokyo metropolitan area,census area data,and national county data from New York in the Unit-ed States,and census area data and county data from London in the United Kingdom.The results show that compared to the classical theoretical models,the machine learning-based analysis framework achieves bet-ter prediction performance and generalization capability.Its average prediction accuracy in different scenar-ios is improved by 96%,98%,and 54%compared to the gravity model,the radiation model,and the PWO model,respectively.Finally,the results of the case studies demonstrate the generalizability of travel dis-tance,population coverage at the origin and destination of travel in explaining population travel patterns,and the importance of the number of travel destination populations is significantly higher than other fac-tors.Meanwhile,the analysis shows that the influence of administrative boundaries will weaken with the increase of the analysis scale,and the mechanism of the origin population size will vary greatly among re-gions with significant economic and cultural differences.These findings provide valuable insights for a deeper understanding of the complexity of travel patterns.
travel distribution modelmobilitymachine learninginfluencing factorstravel patterns
NETL
NSTL
万方数据
