Measuring cell-id trajectory similarity for mobile phone route classification

Route classification based on trajectory data is one of the most essential issues for many location-aware applications. Most existing methods are based on physical locations of the trajectories. However, obtaining physical locations from mobile phones would incur extra cost (e.g. extra energy cost for using GPS). On the other hand, since every active mobile phone is connected to a nearby cell tower, cell-ids (i.e. identifiers of the connected cell towers) could be easily obtained without any additional hardware or network services. In this paper, a cell-id trajectory is a sequence of cell-ids with no regard to physical locations. We address the problem of route classification based on cell-id trajectory data. Specifically, we propose a novel similarity measure which explores the handoff patterns to capture the similarity between cell-id trajectories with no regard to physical locations. Then, based on the cell-id trajectory similarity measure, a clustering algorithm is used to discover potential route patterns from cell-id trajectories, and a nearest-neighbor classification algorithm is used to match current cell-id trajectories to route patterns. The performance of the proposed method is evaluated upon real-world cell-id trajectory dataset. The experimental results showed that our method outperforms state-of-the-art methods on cell-id trajectory clustering and cell-id route classification.