01336820

Component

The probability distribution of duration is critical input for predicting the potential impact of incidents. After analyzing the limitation of naive Bayesian (NB) classifier and unrestricted Bayesian networks (UBN) classifier, tree augmented naive Bayesian (TAN) classifier is chosen to develop a new discrete model for duration prediction. The discrete models divide duration into several intervals, but some time the continuous probability distribution is needed. Therefore the authors also develop a continuous model based on latent Gaussian naive Bayesian (LGNB) classifier, assuming duration fit a lognormal distribution. Both these two models can accommodate incomplete information. These models are calibrated and tested by incident records from the Georgia Department of Transportation. The results show that TAN classifier performs favorably compared to UBN classifier and NB classifier, and LGNB can describe the continuous probability distribution of duration well. According to the evidence sensitivity analysis of LGNB, the four classes of incidents classified by LGNB can be interpreted by the level of severity and complexity. TAN classifier as an extent of NB classifier is still simple but works better, and can replace NB classifier. LGNB classifier combines the Bayesian classifier theory and the continuous probability distribution of duration, can provide more useful information about incident duration and get more application.

01329018

11-0915

English

19p

2011

Transportation Research Board 90th Annual Meeting

DVD

Figures (5) ; References (31) ; Tables (4)

Calibration; Crash records; Crash severity; Incident management; Sensitivity analysis; Time duration; Traffic incidents; Traffic models

Bayesian analysis; Incident duration; Incident severity

Georgia

Highways; Operations and Traffic Management; Planning and Forecasting; I72: Traffic and Transport Planning

Transportation Research Board Annual Meeting 2011 Paper #11-0915

TRIS, TRB

Feb 17 2011 5:37PM