Urban mobility analytics: A deep spatial–temporal product neural network for traveler attributes inference

This study examines the potential of using smart card data in public transit systems to infer attributes of travelers, thereby facilitating a more user-centered public transport service design while reducing the use of expensive and time-consuming travel surveys. This is challenging since travel behaviors vary significantly over the population, space, and time and developing meaningful links between them and traveler attributes are not trivial. To achieve this, we conduct an extensive analysis of spatio-temporal travel behavior patterns using smart card data from the Greater Sydney area (Opal card), and then develop a Hybrid Neural Network to utilize spatial and temporal dependencies in the dataset. In particular, we first empirically analyze passengers’ movements and mobility patterns from both spatial and temporal perspectives and design a set of discriminative features to characterize the patterns. We then propose a deep-learning-based framework to investigate spatial and temporal features in order to infer traveler attributes. The proposed modeling framework consists of two components, i.e., a Product-based Spatial–Temporal Module (PSTM) and an Auto-Encoder-based Compression Module (AECM). PSTM encodes the relationships across a variety of features while AECM derives useful spatial information from a transit stop matrix. The proposed model is tested and evaluated using a large-scale public transport dataset in the Greater Sydney area to infer two attributes of passengers, i.e., the age group and residential area. The experimental results demonstrate the effectiveness of the proposed method against a number of established tools in the literature. The developed techniques can be potentially adapted to other domains where spatio-temporal features are critical, such as commercial/entertainment site selection and urban service planning.