A Mobility Forecasting Framework with Vertical Federated Learning

Abstract

With the prevalence of mobile devices and location-based services, forecasting human mobility has become a critical topic in ubiquitous computing. Existing forecasting approaches usually adopt frameworks with a centralized mobility data holder. However, mobility data typically pertains to independent organizations, introducing two learning challenges. First, since each organization only holds a location domain subset, none can tackle a forecasting model that covers the whole location domain. Second, distributed mobility data compromises the spatio-temporal correlation between locations hindering learning. Hence, reducing the forecasting accuracy. This work proposes a mobility vertical federated forecasting (MVFF) framework that allows the learning process to be jointly conducted over vertically partitioned data belonging to multiple organizations. MVFF enables the forecasting of mobility predictions covering a joint location domain. We evaluate MVFF's performance over two real-world datasets using different spatial and temporal neural network algorithms. Experimental results demonstrate that the two datasets' mean percentage error performance gains are up to 12% and 4% compared to the state-of-the-art, respectively.