Developing a Soft Computing based Decision Tree Approach for Predicting Crashes Severity on Regional Highway Corrid

Developing intelligent and novel methods for crash prevention or reducing crash severity in regional highway corridor is one of the major goals of road safety research. This study presents a comprehensive approach using geospatial information systems and data mining to analyze the severity of highway corridors crashes and identify the most spatial contributing factors. The approach implements Fuzzy Classification and Regression Tree (FCART) on a database of spatial data and four year period accident records in the study corridor (Qazvin-Rasht). The proposed method is tested on the crash data using a 10-fold cross validation process and the results are compared with Classification and Regression Tree (CART) model. The results show that FCART model inducts crash severity better than CART model and its overall accuracy is higher than CART model. Moreover, the sensitivity analysis of FCART model indicates that beside vehicle failure, using seatbelt and weather condition factors, curve and the spatial distribution and prevalence of activities and land uses in the proximity of highway corridors are among the most important factors affecting the severity of injuries and increase opportunities for crash occurrences.