Zero inflated models based on real time traffic characteristics for predicting crash probabilities.

Traffic management Centers (TMCs) at strategic locations in Virginia serve as the depository of the extensive real-time traffic data being currently collected on Interstate Highways through the use of loop detectors installed at regular intervals along these highways. The availability of these data offer the opportunity for their use in these TMCs to control traffic in real time so as to avoid the formation of certain combinations of traffic characteristics that lead to the high probability of crash occurrence.Unfortunately, these data are not currently widely used for this purpose as there no guidelines available to managers of these TMCs that they can use to identify the combination of real time traffic characteristics that are associated with high probabilities of crash occurrence. The development of these guidelines requires first, the identification of those traffic characteristics that lead to high crash occurrence and then the selection of suitable countermeasures that will eliminate or reduce the negative safety impacts of these combinations of traffic characteristics. This paper presents the results obtained from the first part of a study to develop suitable guidelines that can be used by managers of TMCs in Virginia. The paper describes how the crash and real-time traffic characteristics were extracted, the mining of the data and the development of probability models that relate the probability of crash occurrence with the traffic characteristics. For this portion of the study, three data bases (HTRIS, ACCESS, Smart Travel Lab (STL)) were used to extract the data for a period of two years, from July 2003 to July 2005 from segments of I-66 and I-95 located in Northern Virginia Zero Inflated models were developed that relate the probabilities of crash occurrences with traffic characteristics. Predicted probabilities based on the models for crash occurrence closely match those for the actual data, which suggests that these models can be used to identify the safety impact of the different combinations of the traffic characteristics. (A). For the covering abstract of the conference seeE216632.