This paper describes how, because of data and methodology restrictions, it has been difficult for researchers to model the occurrence of accidents. The objective of this paper is two fold. First, model the occurrence of accidents with regard to their nature including: (1) the current incomplete information about accidents; (2) a large number of explanatory variables and their complicated interdependent relationships; (3) the dynamic process of the occurrence of accidents; and (4) the uniqueness of accidents at disaggregate level. Second to uncover accident patterns and features through the analysis. A novel non-parametric methodology- rough sets- combined with a proposed framework which reflects the dynamic process of accidents is adopted. A real world database- 2003 Taiwan single auto-vehicle accidents- is applied to demonstrate the proposed approach. The paper results show that rough sets combined with the proposed framework is feasible to test the existence of accident patterns and their strength. Moreover, the significance of accident features can be shown by their presence percentage on the derived rules.
Abstract