Converting Urban Trips to Multi-Dimensional Signals to Improve Trip Purpose Inference

Smart card data is a rich and available dataset for monitoring and planning public transportation systems. The trip purpose is a valuable attribute of the trip that is missing from the smart card data and many studies have tried to estimate it from other trip attributes. This study infers the trip purpose of public transportation users using a novel method. This method considers both discrete and continuous trip attributes to create a multi-dimensional signal of daily trips for each individual. This signal is an invertible function of independent attributes. In this study, each trip attribute is converted into a dimension of the trip signal. Signals are then clustered to the separated groups using Agglomerative Hierarchical Clustering and Multi-Dimensional Dynamic Time Warping methods. After the clustering, each new passenger is allocated to the closest cluster to predict the purpose of his/her daily trips. Also, a probabilistic approach, which can estimate the percentage of trip purposes for all passengers in a specific area, is used to discover the proportion of each trip purpose among all transportation users in an area. This probabilistic method is shown to increase the total accuracy of purpose inference. This study is implemented on the Household Travel Survey and the smart card data of South East Queensland, Australia. Evaluation outcomes indicated that the proposed method significantly improved the trip purpose inference accuracy.