ATMS Implementation System for Identifying Traffic Conditions Leading to Potential Crashes

Predicting a crash occurrence is the key to traffic safety. Real-time identification of freeway segments with high crash potential is addressed in this paper. For this study, historical crashes and corresponding traffic-surveillance data from loop detectors were gathered from a 36-mi corridor of Interstate 4 for 4 years. Following an exploratory analysis, two types of logistic-regression models (i.e., simple and multivariate) were developed. It was observed that, although the simple models have the advantage of being tolerant in their data requirements, their classification accuracy was inferior to that of the final multivariate model. Hence, the simple models were used to deduce time-space patterns of variation in crash risk while the multivariate model was chosen for final classification of traffic patterns. As a suggested application for the simple models, their output may be used for the preliminary assessment of the crash risk. If there is an indication of high crash risk, then the multivariate model may be employed to explicitly classify the data patterns as leading or not leading to a crash occurrence. A demonstration of this two-stage real-time application strategy, based on simple and multivariate models, is provided in the paper. The output from these model-processing real-time loop-detector data may be utilized by traffic-management authorities for developing proactive traffic-management strategies