Detecting transfer trip patterns in public transport using smart card data

Date

2025-6

Author

Özer, Fehmi Can

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

61
views

33
downloads

Cite This

Public transport (PT) decisions are affected significantly by the potential of and need for trip chaining, which allows transfers across different modes. Thus, detection and analysis of such transfers at PT stops can support further improvements and urban land use changes. Using PT smart card data (SCD), a big data source, can provide valuable insights, but it requires development of GIS-based algorithms that first detect a transfer trip by comparing consecutive boarding data, estimated alighting stops, and transfer duration and characteristics (such as transfers with activity), which is a goal of this study. Analysis of transfer trips in a GIS environment highlights critical PT stops, which serve i) very high transfer rates, ii) direct transfers, and iii) transfers with activities. Further analysis of long-term SCD (i.e., a full month) is used to train machine learning (ML) models (an XGBoost classifier) to predict the transfer nature of a trip based on trip and user features, with strong estimation power (evaluated through accuracy, precision, recall, F1 scores, and confusion matrices). Estimation of spatiotemporal (ST) clustering of user boarding locations using a developed ST-DBSCAN algorithm improved the proposed ML models. Stop-wise accuracy is further mapped categorically to illustrate the spatial justification of the transfer predictions. The numerical results using SCD from Konya, Türkiye (a total of 6 months), showed that the city centre served as practical transfer hubs, as designed, while PT stops in suburban or peripheral neighbourhoods captured fewer transfer events due to less commute-oriented travel.

Subject Keywords

Smart card data, Public transportation, ST-DBSCAN, GIS, Machine learning

URI

https://hdl.handle.net/11511/115082

Collections

Graduate School of Natural and Applied Sciences, Thesis