Predicting Individual Mobility Behavior of Ride-Hailing Service Users Considering Heterogeneity of Trip Purposes

The emergence of on-demand ride-hailing service platforms, such as Uber, Lyft, and Didi, can provide valuable data to understand and model individual mobility behavior. Compared to the previously used data sources such as mobile phone and social media, mobility data extracted from ride-hailing service platforms contain more specific, detailed, and longitudinal information of individual travel mode and visited locations. In this study, using large-scale data extracted from 50,000 ride-hailing service users’ mobility records, we apply a multi-layer hidden Markov model that contains two parts: a trip decision model and a mobility sequence generation model. The trip decision model predicts whether an individual will make a trip for a specific day and the mobility sequence generation model predicts the origin and destination of the next ride-hailing service trip for that individual. The results of the trip decision model indicate that the proposed model can achieve around 65% accuracy for predicting whether the individual would use a ride-hailing service given the contextual information. In addition, the results indicate that the mobility sequence generation model can achieve 71% accuracy for predicting the origin and 67% for predicting the destination of a trip made for commuting purposes. Since individual mobility behavior shows both regularity and uncertainty, we analyze the performance of mobility sequence generation model by investigating the predictability of each mobility sequence. We discovered that model accuracy is proportional to the predictability of individual movement and the model accuracy for a commute-based user is close to the predictability of the person’s movement pattern.