Predicting Traffic Crashes Using Real - Time Traffic Speed Patterns

PREDICTING TRAFFIC CRASHES USING REAL-TIME TRAFFIC SPEED PATTERNS Mohamed Abdel-Aty", Anurag Pande of the recent advances in traffic surveillance technology and concern over there have been very few research efforts between the real-time flow and crash occurrence. This aims at the identification the in the loop detector data. which potentially nr"""'I"'I'" traffic crashes. This would have implications for Advanced (A TMC). ATMCs could then able to predict the potential for and take action to reduce hazard drivers or introducing variable Solution to this research problem involves of rt""'",,,rr.r data. Historical crash and area has been used for this here is the probabilistic neural network (PNN): classifier. The PNN, not statistical basis. The inputs to the model found were logarithms of the coefficient of variation in from three the crash (i.e. station nearest to the crash two stations immediately preceding it in the upstream direction, during 5 minute time slice of 10-15 minutes to crash time. The results showed that about 70% of the crashes on the evaluation could using the classifiers developed """.UUi..,,>, tremendous growth has been observed in advanced (ATMIS). Although the concern over traffic grown in this there were no efforts devoted to prevent crashes these unti I few very recent studies. The conventional to traffic safety analysis has been to establish between the characteristics flow, speed), roadway and environmental of the weather conditions) and driver characteristics and occurrence. The problem with most of the models developed using this that upon measures and MDT or hourly and hence are not sufficient to identifY the "black locations having high probability of crashes), created due to the ambient traffic conditions, using the real-time variables (speed, flow and occupancy) obtained from the loop detectors in an ATMIS environment. Jn this study, the problem of predicting crashes using the loop data has been approached as a classification problem in which we categorize the real-time traffic conditions as measured by loop detectors into either leading or not leading to a crash. The identification of parameters to be used as inputs to the classification algorithm (Probabilistic Neural Network; in this case) is also a part of this study.