Abstract
A crash is often caused by a series of errors and also attributed to a number of categorical explanatory factors. To explore the key rules that determine the most contributing factors to crash severity, this paper develops a novel genetic mining rule (GMR) model, which accounts for the conflict and redundancy of rules mined. To avoid over-mining caused by unevenly distributed data across different types of accidents, identical numbers of A1-type (fatal), A2-type (injury), and A3-type (non-injury) crash cases drawn from 2003-2007 Taiwan's freeway accident investigation reports are used for the analysis. A total of 39 rules are mined which can achieve an overall correct rate of 74.25% in training and 70.79% in validation, respectively, much higher than those yielded by the decision tree model. Travel period, major cause, collision type and journey purpose are found as the four major contributory factors to crash severity in this study.
