The traditional method of identifying road accident hot zones is through the examination of accident frequency and nature, which sometimes, can be subjective and inaccurate. To overcome the limitation of the traditional method, researchers have applied Geographic Information System (GIS) approaches to identify and visualise road traffic accidences in real-time. However, these approaches still treat accidences as occasional and discrete events and can’t support accurate analysis and prediction of accidences at some point. This paper takes the spatial autocorrelation nature of accidents (i.e., the interdependence of accident data and the relationship between accidents and the space) into account and proposes an innovative spatial-autocorrelation-based method to identify freeway accident hot zones. Based on the spatial autocorrelation and mathematical statistics, this method constructs a point-line connectivity network to realise the space localisation and validation of accidents. Combined with GIS approaches, our approach can also automatically identify and visualise accident-prone areas. At the moment, the approach has been tentatively applied in a highway in China. The result demonstrates the algorithm behind the approach can effectively convert accident data into spatial data, cluster accident hot zones of any length and predict the whereabouts of likely accidents in the future. In conclusion, the innovation can be reflected from the approach robustness and accuracy in accident-prone road section identification.

CICTP