A Study to Predict the Traffic Accident Severity Level Applying Neural Network at the Signalized Intersections

The number of signalized intersection accidents were about 21% of total traffic accidents in 2001 year, and there has been an increasing number of traffic accidents at intersection since 1990s. Although many studies for a safety evaluation methods at intersections have been progressed, most of these studies have not used general data which applied to all intersections. Therefore, this study developed two prediction models of the traffic accident severity levels using traffic conflict data that were collected at all of the intersections. These prediction models of the accident severity levels which are made up a multiple regression analysis and a neural network model that used the error backpropagation algorithm were made of using the gap between vehicles and speed, accident severity levels of existing traffic accident data at signalized intersection. When each of these two models predicted the twenty existing traffic accident severity levels, the prediction ability of neural network model was better than that of the multiple regression model, The collected data from the field which are gap between vehicles and speed of traffic conflict data at intersection were applied to the prediction models, and these models predicted that if the conflict connected a traffic accident, it will more or less raise the traffic accident severity level. When we evaluate the safety at intersection in case of lack of traffic accident data, we are able to collect such traffic conflict data from the field. Accordingly as these data is applied to the models, we can decide the risk-level at intersections which are based on the prediction of traffic accident severity levels.