Title: A multi-feature fusion method for travel demand prediction based on federated learning

Authors: Kangyi Du; Zheng Huo; Xiangshai Zhang

Addresses: Management Science and Information Engineering School, Hebei University of Economics and Business, Shijiazhuang, Hebei Province, China ' Management Science and Information Engineering School, Hebei University of Economics and Business, Shijiazhuang, Hebei Province, China ' Management Science and Information Engineering School, Hebei University of Economics and Business, Shijiazhuang, Hebei Province, China

Abstract: Shared mobility has become an emerging transportation mode in smart cities. However, balanced utilisation of shared bicycles faces challenges due to uneven spatiotemporal user demand, asymmetry in station numbers, and decentralised data storage by service providers. To address these issues, we propose Fed-STID, a federated spatio-temporal demand prediction framework that integrates multi-source features including trip history, temporal and spatial dependencies, and weather conditions. Built on a federated multi-layer perceptron, Fed-STID enables accurate prediction on horizontally partitioned data while preserving privacy. Experiments on real-world bike-sharing datasets demonstrate that Fed-STID consistently outperforms mainstream baselines, achieving higher accuracy and lower mean squared error. These results highlight the effectiveness and scalability of our approach for large-scale smart mobility systems.

Keywords: shared mobility; federated learning; spatiotemporal features; multi-layer perceptron.

DOI: 10.1504/IJCSE.2026.155104

International Journal of Computational Science and Engineering, 2026 Vol.29 No.4, pp.372 - 383

Received: 09 Apr 2025
Accepted: 12 Jan 2026
Published online: 27 Jul 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article