Search papers, labs, and topics across Lattice.
The paper introduces Bikelution, a federated learning approach using gradient-boosted trees for bike-sharing demand forecasting. This method addresses privacy and bandwidth limitations associated with centralized machine learning while maintaining forecast accuracy. Experimental results on three real-world datasets demonstrate that Bikelution achieves performance comparable to centralized methods and surpasses existing state-of-the-art techniques.
Federated learning can match centralized ML performance in bike-sharing demand forecasting, opening the door to privacy-preserving urban mobility solutions.
The rapid growth of dockless bike-sharing systems has generated massive spatio-temporal datasets useful for fleet allocation, congestion reduction, and sustainable mobility. Bike demand, however, depends on several external factors, making traditional time-series models insufficient. Centralized Machine Learning (CML) yields high-accuracy forecasts but raises privacy and bandwidth issues when data are distributed across edge devices. To overcome these limitations, we propose Bikelution, an efficient Federated Learning (FL) solution based on gradient-boosted trees that preserves privacy while delivering accurate mid-term demand forecasts up to six hours ahead. Experiments on three real-world BSS datasets show that Bikelution is comparable to its CML-based variant and outperforms the current state-of-the-art. The results highlight the feasibility of privacy-aware demand forecasting and outline the trade-offs between FL and CML approaches.
