Trip chain extraction using smartphone-collected trajectory data

Travel trajectories of individuals, collected by smartphone sensors, carry substantial trip chain information on advanced urban transportation operation. Automatic information extraction plays a critical role in intelligent transportation data collection and analysis, with higher efficiency and accuracy. This research proposes an entire trip chain extraction procedure, including integrated activity/trip segmentation and travel-mode and activity-type recognition components. Specifically, a unique sliding-window Euclidean distance method is proposed to segment trajectories into travel and activity segments. Four alternative black-box pattern recognition (PR) methods, including decision trees (DT), multilayer perceptrons (MLP), radial basis function neural networks (RBFNN), and support vector machines (SVM), are used to classify travel-mode and activity type. An F-Score-based feature-selecting rule is further introduced to improve the performance of the travel-mode/activity-type classifiers. The proposed extraction procedure is demonstrated using trajectory data collected from volunteers. Test results show that the proposed sliding-window Euclidean distance segmentation approach has error rates as low as 2–4%; recognition tests suggest that SVM performs the best in travel-mode recognition and MLP performs the best in activity-type recognition; feature selection tests reveal that the methods with feature selection achieve higher accuracy than those with total features. Overall, the proposed trip chain extraction procedure achieves an average completion rate above 80%.