Identifying the travel patterns of on-demand taxi trips through inferred trip purposes

Abstract

Activity-based modeling has become the backbone behind transportation planning, and trip purposes that can be inferred from large GPS datasets of different travel means are paving the path to augmenting its accuracy. In this context, the trip purpose inference problem has emerged since the GPS is unable to capture the trip purposes explicitly. This problem has not been thoroughly addressed in developing countries despite the fact that the applicability of a trip purpose inference model extremely depends on the land use context. Hence, this study attempted to ameliorate an accurate model (base model) for a chosen study area in Colombo District, Sri Lanka using the on-demand taxi trips data. Point of Interest (POI) data is often accompanied by the purpose inference models as it provides a complete insight into the land use around origin and destinations. In the pursuit of the main objective of the study, machine-learning-based text classification was tested to improve the number of informative POIs and its outcome indicated that the Support Vector Machine (SVM) classifier can be utilized effectively and efficiently. The designed trip purpose inference model referring to a base model is proposed as a three-layer trip purpose inference framework in which a method to impute purpose based on trip regularities and the method to identify residential trips was included as two layers before using the base model. The validity of the proposed model was evaluated with the assistance of household travel survey data and with respect to the purpose proportions and division level R-square. Furthermore, travel patterns of the on-demand taxi trips were assessed in terms of temporal regularity, trip lengths, and spatial dynamics. It is recommended to conduct further studies to assess the applicability of other parameters such as trip origin context and trip time with the assistance of unsupervised learning methods.