Modeling Fine-Grained Human Mobility on Cellular Networks

Cellular network data has been proved as one of the most promising ways to understand large-scale human mobility due to its high penetration of cellphones and low collection cost. Most existing mobility models driven by cellular network data are based on either CDR (Call Detail Records) or data connection records. However, estimated mobility is biased with coarse granularities due to the insufficient data quality. Mobility modeling on cellular networks always suffer from the sparse spatial-temporal observations since user locations are recorded with cellphone activities. In this paper, to solve the issue, we design a system named FineCell to model fine-grained human mobility based on sparse cellular network data. The key challenge we address in FineCell is to achieve fine-grained mobility modeling with sparse cellular network data. In contrast to the existing works on human mobility, the novelty of the FineCell is to infer missing spatial and temporal observations caused by sensing gaps in cellular networks. More importantly, we evaluate FineCell with large-scale fine-grained ground truth data from the same cellular network. The evaluation results show FineCell achieve 9.8% lower error compared with state-of-the-art models.