Travel pattern-based bus trip origin-destination estimation using smart card data

Abstract

Smart card data are widely used in generating the origin and destination (O–D) matrix for public transit, which contains important information for transportation planning and operation. However, the generation of the O–D matrix is limited by the smart card data information that includes the boarding (origin) information without the alighting (destination) information. To solve this problem, trip chain methods have been proposed, thereby greatly contributing in estimating the destination using the smart card data. Nevertheless, unlinked trips, that is, trips with unknown destinations, are a persisting issue. The purpose of this study is to develop a method for estimating the destination of unlinked trips, in which trip chain methods cannot be applied, using temporal travel patterns and historical boarding records of the passengers based on long-term smart card data. The passengers were clustered by k-means clustering, and the time-of-day travel patterns were estimated for each cluster using a Gaussian mixture model. The travel patterns were formulated to estimate the destination of the passengers from the smart card data. The proposed method was verified using the 2018 smart card data collected in Sejong City, South Korea. The existing trip chain method matched the destinations of 60.0% of the total trips, whereas the proposed method improved the matching to 74.9% by additionally matching the destinations of 37.2% of the unlinked trips.

Introducing Autonomous Shuttle Services Based on Travel Patterns for the Elderly

“…As the partitioning clustering algorithm has a linear time complexity with respect to the data size, it can efciently process large datasets [38]. Consequently, it has been applied to various research felds and is widely used in studies involving large smart-card datasets [39,40]. Te most typical partitioning clustering algorithm, the k-means algorithm, rapidly determines cluster membership by minimizing the distance between the central point and the other data points in the cluster.…”

“…In contrast, the Gaussian mixture model (GMM) can identify clusters with diferent sizes or densities and more general shapes; thus, it is more fexible than that of the kmeans clustering [43,44]. In many cases, the GMM algorithm shows a higher classifcation accuracy than that of the k-means clustering [42,45] and model-based clustering methods are suitable for modeling similar travel patterns [36,40]. We explored the characteristics of each algorithm and compared them to the dataset used in this study.…”

Investigating the potential of aggregated mobility indices for inferring public transport ridership changes

“…By contrast, cities that have already adopted automated fare collection (AFC) schemes have had the advantage of analysing their PT demand information from smart cards and digital transactions [7][8][9]. However, some limitations on the fare collection system may affect the quality of these data [10]. For example, ridership data may be lower than the actual one when ticketing are missing or incomplete, such as in the cases of ticket-free riding days or when there are special periods where fare evasion is potentially higher.…”