Learning Transportation Mode Choice for Context-Aware Services with Directed-Graph-Guided Fused Lasso from GPS Trajectory Data

Mobility profiles of users play a crucial role in a wide range of context-aware computing and services. Travel mode choice, as a representative feature of mobility profiles, is one of the important components in travel demand and future planning of transportation systems. Transportation mode choice has been widely studied based on the random utility model and decision making methods which haven't considered the correlation among features influencing transportation mode choice. This paper presents a data driven model to analyze transportation mode choice given transportation information. The contributions of this paper lie in the following two aspects. On one hand, we propose a travel mode choice model considering the correlation among influencing features of mode. And the relevant features related to the mode choice are redefined and considered to improve the final efficiency and effectiveness. On the other hand, we propose a directed-graph-guided fused lasso method to depict the correlation rules among features. The lasso method can reduce the redundant information to improve the speed of convergence and accuracy of analysis. Three different models namely standard lasso, graph-guided fused lasso and spatio-functionally weighted regression based models, are compared with our model and tested with the GPS trajectory data in Beijing. As a result, we achieved better performance than other compared models.