Asad J. Khattak, Joseph L. Schofer, Mu-Han Wang
1995 I V H S Journal  
The objective of this study is to develop a methodology for incident duration prediction. First, we develop an understanding of factors that influence incident duration. Then, we use a series of truncated regression models to predict incident duration. The models account for the fact that incident information at a Traffic Operations Center is acquired over the life of the incident. The implications of this simple methodology for incident duration prediction are discussed. Summary The prediction
more » ... of incident durations can facilitate incident management and support traveler decisions. This paper develops a procedure for predicting incident durations. First, the causal and non-causal factors which influence incident durations are conceptualized. These include operational characteristics such as response times and whether a heavy wrecker was used, incident characteristics such as injuries and number of vehicles involved and environmental conditions such as weather and visibility. Specific hypotheses are tested by developing truncated regression models of incident duration using data provided by the Illinois Department of Transportation (IDOT) on Chicago area freeways. The models show that incident durations are longer when the response times are higher, the incident information is not disseminated through the public media, there are severe injuries, trucks are involved in the incident, there is heavy loading in the truck, State property is damaged, and the weather is bad. The most important variables in incident duration prediction were incident characteristics and the consequent emergency response actions. A time sequential methodology is developed to predict the incident durations as information about the incident is acquired in a Traffic Operations Center or TOC. Initially, after an incident is detected, information at a TOC is often acquired at a high rate, then information acquisition levels off and toward the end of an incident the acquired information may decay. Accordingly, the incident duration models grow in terms of their explanatory variables at first, then they are sustained during the middle stages and begin shrinking toward the end when information starts decaying. The estimated time sequential models remain statistically significant even after they have shrunk to a few remaining incidents. One purpose of this prediction methodology is to demonstrate the trade-off between early and accurate incident duration prediction. For the models to be operational, we suggest estimating similar models with more local data. Abstract summary Acknowledgement vi
doi:10.1080/10248079508903820 fatcat:ug4kdxp575hylegotrasyye6vy