Large-Scale Spatiotemporal Density Smoothing with the Graph-fused Elastic Net: Application to Ride-sourcing Driver Productivity Analysis
share
Ride-sourcing or transportation network companies (TNCs) provide on-demand transportation service for compensation, connecting drivers of personal vehicles with passengers through the use of smartphone applications. This article considers the problem of estimating the probability distribution of the productivity of a driver as a function of space and time. We study data consisting of more than 1 million ride-sourcing trips in Austin, Texas, which are scattered throughout a large graph of 223k vertices, where each vertex represents a traffic analysis zone (TAZ) at a specific hour of the week. We extend existing methods for spatial density smoothing on very large general graphs to the spatiotemporal setting. Our proposed model allows for distinct spatial and temporal dynamics, including different degrees of smoothness, and it appropriately handles vertices with missing data, which in our case arise from a fine discretization over the time dimension. Core to our method is an extension of the Graph-Fused Lasso that we refer to as the Graph-fused Elastic Net (GFEN).
READ FULL TEXT