Social media data has emerged as an innovative data source for traffic analysis. In this paper, we evaluate the effectiveness of including Twitter data into the Origin-Destination (OD) trip estimation. 1.3 million of geo-tagged tweets in the Greater Sydney Area for more than two months are collected, and information such as Twitter OD trips, the number of friends and followers of Twitter users are extracted as the independent variables in the OD trip regression model. The Random Forest regression technique is applied to develop the OD trip regression. The performance of the models considering Twitter data and not including Twitter data are compared via 10-fold cross-validation method. The results indicate that the accuracy and stability of the RF regression model can be improved if we consider Twitter data in the independent variables. Inspired from this finding, we conclude that social media data can be an effective data source to improve the prediction of traditional travel demand models. The regression results at the suburb level also suggest that the heterogeneity of socio-demographic features across suburbs will affect the model performance. To further improve the prediction, it is necessary to categorize suburbs into groups based on socio-demographic characteristics such as population density and distance to city center, and develop a separate OD trip regression model for each group.

Keywords