AIDC project number:
Guohui Zhang
US DOT UTC: CSET
It is crucial to examine the characteristics and attributes of traffic crashes in Rural, Isolated, Tribal, or Indigenous (RITI) communities using statistical and data-driven methods. However, traditional crash data analysis faces challenges due to unobserved heterogeneities and temporal instability. To address these issues, a fusion convolutional neural network with random term (FCNN-R) model is developed for driver injury severity analysis. The proposed model consists of a set of sub-neural networks (sub-NNs) and a multi-layer convolutional neural network (CNN). Seven-year (2010-2016) single-vehicle crash data is applied. The proposed model outperformed other five typical approaches in the predictability comparison. In addition, unobserved heterogeneity, which has been recognized as a critical issue in crash frequency modelling, generates from multiple sources, including observable and unobservable factors, space and time instability, crash severities, etc. In this project, hierarchical Bayesian random parameters models with various spatiotemporal interactions are further developed to address as well. Selected for analysis are the yearly county-level alcohol/drug impaired-driving related crash counts data of three different injury severities including minor injury, major injury, and fatal injury in Idaho from 2010 to 2015. Significant temporal and spatial heterogenous effects are detected in all three crash severities. These empirical results support the incorporation of temporal and spatial heterogeneity in random parameters models.
