Spatial decision tree-application to traffic risk analysis

Spatial data mining fulfils the real needs of many geomatic applications. It allows taking advantage of the growing availability of geographically referenced data and their potential richness. This includes the spatial analysis of risk such as epidemic risk or traffic accident risk in the road network. This work deals with the method of decision trees for spatial data classification. This method differs from conventional decision trees by taking account of implicit spatial relationships in addition to other object attributes. Our approach consists in materializing those spatial relationships leading to treat them as normal attributes. Then, any conventional decision tree building method could be applied providing a spatial decision tree. Compared to existent approaches, this one is more flexible because no specific algorithm is imposed. Moreover, it considers the organization in several thematic layers that is characteristic of geographical data by distinguishing the intra theme and the inter theme relations (such as the road section contiguity or the proximity between road sections and schools). This method has been tested in the framework of traffic risk analysis