01626393

Component

Compared with the large amount of research using various ways of addressing serial correlations among crash data, there is little research dedicated to the evaluation of the different temporal treatments on modeling performance. To add to the current literature the much-needed research, this study first proposed a new method which combines the strengths of time-varying coefficients and autoregressive process, and then compared its performance with seven other temporal models used in the past. Ten years of crash data and other covariates associated with traffic analysis zones in the City of Irvine, California were employed. Bayesian hierarchical approach was employed to account for the structural heterogeneities. The comparisons were conducted for assessment of goodness-of-fit, accuracy of crash estimation, and relative performance of site ranking. The modeling results indicated that the proposed model appeared to have the best fit with actual crash data and a relatively lower complexity than other competing models. Longitudinal and cross sectional validations using RSS (Residual Sum of Squares) demonstrated that the proposed model had very significant superiority at crash prediction with an RSS score three times smaller than the worst performing model. The site ranking evaluation established that there is a statistically significant correlation between site ranking performance and modeling performance.

This paper was sponsored by TRB committee ANB20 Standing Committee on Safety Data, Analysis and Evaluation.

01618707

17-05548

English

20p

2017

Transportation Research Board 96th Annual Meeting

Digital/other

Figures; Maps; References (54) ; Tables

Crash data; Crash rates; Mathematical prediction; Time intervals; Time periods

Irvine (California)

Highways; Safety and Human Factors

Transportation Research Board Annual Meeting 2017 Paper #17-05548

TRIS, TRB, ATRI

Dec 8 2016 12:12PM