Technical gazette, Vol. 32 No. 2, 2025.
Original scientific paper
https://doi.org/10.17559/TV-20240601001729
VFed-PU: Identifying Containers with Potential to be Shipped by Rail from Ports with Privacy Protection
Lei Huang
; School of Economics and Management, Beijing Jiaotong University, 10044, Beijing, PR China
*
Deyou Jiang
; School of Economics and Management, Beijing Jiaotong University, 10044, Beijing, PR China
Xiong Zhang
; School of Economics and Management, Beijing Jiaotong University, 10044, Beijing, PR China
Ying Wang
; School of Economics and Management, Beijing Jiaotong University, 10044, Beijing, PR China
Tianyang Bai
; China Waterborne Transport Research Institute, 100088, Beijing, PR China
* Corresponding author.
Abstract
Facing challenges in the global container shipping market and strict data protection laws like GDPR and CCPA/CCPR, the sea-rail intermodal transportation sector urgently needs better freight demand forecasting. This study develops a micro-level transportation demand forecasting model tailored for the sea-rail intermodal sector, emphasizing data privacy and accurate prediction of port demand for container shipments by rail, which is crucial for effective railway planning and marketing. We introduce a novel framework, VFed-PU, which combines Vertical Federated Learning with Positive and Unlabeled Learning. This model tackles issues such as limited labeled data, data imbalance, and selection bias using a new method called ImbalancednnPUSB. VFed-PU ensures data privacy by transferring only data representations rather than the original data during model training, safeguarding sensitive information among different parties. Extensive experiments demonstrate that VFed-PU outperforms state-of-the-art algorithms in predicting port demand for container shipments, achieving a recall rate of approximately 90%. This framework not only enhances prediction accuracy and preserves data privacy but also supports strategic railway planning and marketing efforts. The study highlights the importance of data privacy in transportation planning, especially under stringent data protection regulations, and contributes significantly to the field by addressing both forecasting performance and privacy concerns.
Keywords
freight demand forecasting; positive and unlabeled learning; sea-rail intermodal transportation; vertical federated learning
Hrčak ID:
328567
URI
Publication date:
27.2.2025.
Visits: 874 *