Generating Time-Varying Road Network Data Using Sparse Trajectories

While research on time-varying graphs has attracted recent attention, the research community has limited or no access to real datasets to develop effective algorithms and systems. Using noisy and sparse GPS traces from vehicles, we develop a timevarying road network data set where edge weights differ over time. We present our methodology and share this dataset, along with a graph manipulation tool. We estimate the traffic conditions using the sparse GPS data available by characterizing the sparsity issues and assessing the properties of travel sequence data frequency domain. We develop interpolation methods to complete the sparse data into a complete graph dataset with realistic time-varying edge values. We evaluate the performance of timevarying and static shortest path solutions over the generated dynamic road network. The shortest paths using the dynamic graph produce very different results than the static version. We provide an independent Java API and a graph database to analyze and manipulate the generated time-varying graph data easily, not requiring any knowledge about the inners of the graph database system. We expect our solution to support researchers to pursue problems of time-varying graphs in terms of theoretical, algorithmic, and systems aspects. The data and Java API are available at: http://elif.eser.bilkent.edu.tr/roadnetwork.