Authors: Woon Kim, Suhasini Natarajan, and Gang-Len Chang
Conference: The 11th International IEEE Conference on Intelligent Transportation System, October 2008 in Beijing, China.
This paper presents a methodology for developinga model to identify the variables influencing incident duration to estimate and predict incident duration in the state of Maryland. The incident information from years 2003 to 2005 from the Maryland State Highway (MDSHA) database was used for model development, and year 2006 for model validation. Classification Trees (CT) were employed for a preliminary analysis to understand the influence of the variables associated with an incident. Based on the findings from CT, this study employed the Rule-Based Tree Model(RBTM) to develop the primary prediction model. The overall confidence for the estimated model was over 80% with several remarkable findings regarding the association between the identified factors and incident duration. Although the estimated results from RBTM were quite acceptable, in cases where RBTM did not provide incident duration within a desirable short range, a discrete choice model was developed as a supplemental model. It is deduced that supplemental models along with better quality database are required to improve the prediction accuracy of the duration of a detected incident.