B0613
Title: Scalable warped directional traffic network models for traffic accident data
Authors: Philip White - Brigham Young University (United States) [presenting]
Abstract: Traffic accident data from the Utah Department of Transportation are considered. Specifically, we consider six years of traffic accidents from I-15, the most heavily trafficked highway in Utah, USA. In total, there are 48,704 traffic accidents. For these data, we propose a scalable model to identify roadway characteristics associated with more traffic accidents, learn dynamics in accident patterns, capture nonstationary patterns present in the data, and forecast future accident patterns. We present a scalable framework for approximating log-Gaussian Cox processes. In addition, we use spatial warping to capture non-stationary patterns in traffic accident data. We also include dynamics and direction in the regression coefficients and the spatial model to capture year-to-year and directional variation in accident patterns. We compare various model specifications. For the best model, discuss the results of this model in this dataset.