Analyzing heterogeneous accident data from the perspective of accident occurrence

Clustering and classification approaches have been commonly applied in reducing the heterogeneity in accident data. As part of an effort to understand the features of the heterogeneity, this study assessed accident data from the perspective of accident occurrences. Using the rule-based classification method, rough set theory, rules were derived which consisted of indispensable factors to certain accident outcomes and reflected the process of accident occurrences. The occurring frequency of each derived rule was then adopted as the basis for grouping accidents for further analyses. Empirical results showed that rules with high occurring frequencies were largely related to drivers with high-risk characteristics. On the other hand, road facilities played a key role in rules with low-occurring frequencies. The distinctive features indicated the essential differences between the frequently repeated and the sparsely unique processes of accident occurrences. This suggests that the heterogeneity of accident data is not limited to one single factor, such as age, gender or area. Thus, the proposed approach, which takes the process of accident occurrences into consideration, can be a potential alternative to more comprehensively analyze the heterogeneity in accident data.