Comparison of Fuzzy and Neural Classifiers for Road Accidents Analysis

This paper investigates the classification of road accidents using neural networks and fuzzy classification techniques. Accident correctability by road improvements has been advocated as an important criterion in the identification of accident-prone locations. Sayed et al. (1995) described a method to identify accident correctability using a fuzzy classification algorithm (fuzzy K-nearest neighbors). The method used safety experts' knowledge to classify accidents according to their contributing factors and causes into the three road-system components (the vehicle, the driver, and the road environment). The membership of each accident in the road-environment component was used as a measure of accident correctability by road improvements. The advantage of this method is its ability to identify the accident-prone locations that are most promising to be treated by road improvements, thus increasing the potential effectiveness of safety improvement projects. This paper introduces a further refinement to the classification process: artificial neural networks. A comparative evaluation of the fuzzy K-nearest neighbors and multilayer feed-forward back-propagation neural network classifiers is carried out. The theoretical backgrounds of both classifiers are presented and their relative advantages are discussed. Results showed that artificial neural networks can produce slightly more accurate results and achieve higher computational efficiency than fuzzy classification.