Abstract for presentation (Poster or Podium)
Data Sensing and Analytics
Ian J. Bonam, MS
Master's Student
University of Georgia
Cumming, GA, United States
Ian J. Bonam, MS
Master's Student
University of Georgia
Cumming, GA, United States
Jidong J. Yang, PhD
Associate Professor
University of Georgia
Athens, GA, United States
Jidong J. Yang, PhD
Associate Professor
University of Georgia
Athens, GA, United States
Jidong J. Yang, PhD
Associate Professor
University of Georgia
Athens, GA, United States
Current traffic safety analysis relies heavily on tabular data. Recent advancements have introduced neural oblivious decision ensembles (NODE), a novel approach that outperforms conventional gradient boosted decision trees in handling various tabular datasets. Taking advantage of the continuously expanding high-dimensional tabular data collected through modern traveler information systems, we explore the capability of NODE for predictive crash modeling. This unique architecture extends the capabilities of tree ensembles, enabling the learning of multilayer hierarchical representations with increased depth. In this study, we apply this advanced structure to address the complexity inherent in crash analysis and modeling. Our focus is on utilizing multisource high-resolution data collected through the Georgia 511 system. For model interpretation, we employ SHAP (Shapley Additive exPlanations) and compare its performance with conventional methods to illustrate its effectiveness.