Nested Sorting and Clustering for Visualization of Accidents in Highway–Rail Grade Crossings

Abstract

Safety at railroad grade crossings is a critical issue for highway and rail networks. Relevant databases can be analyzed to find the causes of or factors contributing to crossing accidents so that appropriate counter-measures can be applied. Manually establishing the order of the contributing factors of the accidents and extracting useful information from accident databases are not feasible because of the enormous number of possible permutations of contributing factors. This paper presents a new automated method for sorting and clustering accident attributes to identify and visualize trends in the accident databases. The method is called modified nested sorting and crossing cluster (M+C). The method creates a dynamic tree visualization that highlights attributes resulting in the greatest accident concentration along a tree branch, uncovering the most common nested accident factors. This approach is a significant improvement over static methods that rely on a fixed hierarchy of attributes. With the M+C method, a unique hierarchy of the attributes can be determined for a single crossing or for a set of crossings. This approach is completely data driven and is suitable for corridors and large groups of accidents that are otherwise difficult to analyze. The method is illustrated for single crossings and a corridor with several crossings identified from the FRA online database. Absolute sorting and nested sorting are discussed, and the evolution of M+C is presented. The M+C method is useful for assessing many single grade crossings or crossings that are along a corridor or within a region.