Using Data Mining Techniques to Understand Collision Processes

摘要: In order to improve road safety, it is necessary better understand collision processes, i.e. the chains of events that lead collisions. Among most important benefits, more efficient countermeasures can be found target causes and factors known This would also help develop reliable surrogate safety measures based on traffic without a have stronger links paper reports first phase project relying microscopic data extracted from video sensors mining techniques identify patterns in dataset with collision. approach demonstrated collected Kentucky 295 events, constituted 213 conflicts 82 Using k-medoid algorithm, its clustering yields three groups distinct characteristics, especially related speed type, or lack, evasive action. The attributes determine event outcome (collision not) are identified through logit model.