Abstract

Spatial data mining, i.e., discovery of interesting, implicit knowledge in spatial databases, is a highly demanding field because very large amounts of spatial data have been collected in various applications, ranging from remote sensing, to geographical information systems (GIS), computer cartography, environmental assessment and planning, etc. In this paper, an efficient method for building decision trees for the classification of objects stored in geographic information databases is proposed and studied. Our approach to spatial classification is based on both (1) non-spatial properties of the classified objects and (2) attributes, pred- icates and functions describing spatial relations between classified objects and other features located in the spatial proximity of the classified objects. Several optimization techniques are explored, including a two-step spatial computation technique, use of spatial-join indices, etc. We implemented the algorithm and conducted experiments that showed the effectiveness of the proposed method.

Citations