Spatio-Temporal Pattern Discovery Methods in Traffic Data: An Overview

As distributed sensor systems expand within extensive urban transportation networks, the substantial volume of traffic data has attracted broad interest from both industrial and academic communities. With the availability of traffic data, the spatio-temporal pattern discovery in traffic data (STPDT) has emerged as a prominent topic. It explores how spatial traffic features evolve over time within a low-dimensional space, thereby discovering desired patterns demonstrating the road network’s normal and anomalous traffic states. Based on a thorough investigation into the state-of-the-art STPDT, this survey undertakes the following efforts: (a) categorizing recent advancements in STPDT approaches; (b) providing widely used baseline traffic datasets; (c) comparing the performance of different commonly adopted models on several real-world public datasets; and (d) identifying unique research opportunities and future directions for STPDT. In doing so, this survey seeks to deliver an in-depth and systematic review of current STPDT methods from the perspectives of temporal and spatial dependencies, thereby facilitating future research on this emerging and vital issue.