Travel Links Prediction in Shared Mobility Networks Using Graph Neural Network Models

The emerging sharing mobility systems are gaining increasing popularity because of the significant economical and environmental benefits. To facilitate the operation of sharing mobility systems, many studies are conducted to analyze and predict users’ travel behaviors. However, most research focuses on investigating every station’s usage and demand; therefore, insight into the user behavior and travel demand between stations from origin to destination is little known. Aiming to better understand the factors that would influence origin-destination travel demand, we present a complex network-based approach to predicting the travel demand between stations (e.g., whether two stations have sufficient trips to form a strong connection in a month) in sharing mobility systems. Particularly, in this study, we are interested in knowing whether local network information (e.g., the neighboring station’s features of a station and its surrounding points of interest (POI), such as banks, schools, etc.) would influence the formation of a strong connection or not. If so, to what extent do such factors play a role in it. To answer this question, we adopt Graph Neural Network (GNN), in which the concept of network embedding can capture and quantify the effect of local network structures. The results are compared with the regular artificial neural network (ANN) model without network embedding. This study is demonstrated using the bike sharing system, Divvy Bike in Chicago, as an example. We observe that the GNN prediction gains up to 9% higher performance than that of the ANN model. This implies that the local network information contributes to the formation of sharing mobility network. Moreover, it is found that when predicting the following year’s network, the model that employs the node embedding obtained from the previous year’s network outperforms the model with the node embedding obtained from the ANN predicted networks.