Trip distribution modeling with Twitter data

Abstract

Integrating both traditional and social media data, this study compares the performance of gravity, neural network, and random forest models of commuting trip distribution in New York City. Trip distribution modeling has primarily employed traditional data sources and classical methods such as the gravity. However, with the emergence of social media during the past decade, the potential for integrating traditional and social media data while utilizing new techniques has been identified. Our findings indicate that the random forest model outperforms the traditional gravity and neural network models. Population, distance, number of Twitter users, and employment were identified as the four most influential predictors of trip distibution by the random forest model. While Twitter flows did not enhance the models' performance, the importance of the number of Twitter users at work destinations implies the potential for using social media data in travel demand modeling to improve the predictive power and accuracy.