ANALYSING TRAFFIC SAFETY FROM A CASE-BASED REASONING PERSPECTIVE

This thesis contains a study of how to use partial matching and qualitative reasoning methods to provide computer-based support for assisting municipal traffic engineers in finding solutions to traffic safety problems. The study was motivated by needs in the application domain, where it was conjectured that decisions regarding accident prevention measures could benefit from easy access to previous relevant experience. The specific domain chosen for the study was urban intersections and conflicts between protected and unprotected road users. Case-based reasoning, CBR, offers a way to represent previous relevant experience, which can be retrieved and used by analogical reasoning. In the study, a series of experiments was performed in order to study the utility of the CBR approach in the traffic safety domain and to assess the feasibility of building systems assisting a traffic engineer in the process of coming up with unconventional solutions based on analogies with other cases. The study indicates that a useful model for representing cases may involve three different structures, namely a functional abstraction hierarchy, a physical aggregation hierarchy and a casual state-transformation accident model. The purpose of these models is to allow partial matching between a posed question (current safety problem) and stored cases (previous experience). The thesis accounts for the experience from test implementations using real accident reports and describes the proposed approach to represent safety-related cases based on system and accident theory. (A)