This thesis describes the development of a highway safety expert system. The objective of the system is to provide highway safety officials with an efficient and reliable tool to identify accident prone locations and then quickly and reliably advise on the appropriate countermeasure(s) based on an analysis of the accident and roadway environment data. The main advantage of the system is its ability to process a large amount of accident data, separating locations which are most promising to be treated by engineering measures and providing advice on the countermeasures and their expected effectiveness. The system also provides an enhancement to many of the techniques currently used in highway safety improvement programs including two new methods for identifying accident prone locations. The nature of traffic safety problems which are ill-structured, poorly understood, and lack explicit algorithms makes it well suited to the expert system approach. The system consists of three basic phases:​ Detection;​ Diagnosis;​ and Remedy. The three phases comprise the main components of highway safety improvement projects. This thesis describes the development of both the detection and the diagnosis phases. The issues which may arise during the development of the remedy phase are also discussed.

The detection phase consists of two components:​ The first, the modified black spot, considers that, from a highway agency’s perspective, accidents which occur due to road related factors should have greater influence in identifying accident prone locations than those which occur due to driver or vehicle related factors. The basic idea is to classify accidents according to their patterns and contributing factors into one of the three groups of the highway system (the driver, the vehicle, and the road environment). A fuzzy pattern recognition algorithm is used for the classification process. Locations are then identified as accident prone if they exhibit a significant number of correctable (e.g. road related) accidents. The second component of the detection phase, the countermeasure-based program, attempts to identify locations which can be cost effectively treated irrespective of their total number of accidents. The approach reverses the traditional process by first identifying main accident patterns that can be targeted by specific countermeasures and then searching for locations which have over-representation of these patterns. The approach utilizes the Empirical Bayes technique for the identification process. Case studies are used to demonstrate the usefulness of the two programs. In the diagnosis phase, a prototype knowledge-based system is developed to identify the causes and the contributing factors of accidents at the locations identified in the detection phase. The output of the diagnosis phase is a set of applicable countermeasures for each accident prone location and the degree of belief in each countermeasure. The prototype knowledge-based system was validated using several case studies which demonstrated satisfactory results.

Finally, several recommendations for further research in selected areas to further enhance the system are introduced.