A BAYESIAN NETWORK APPROACH TO MINE SPATIAL DATA CUBE

摘要

Spatial data mining is an extension of data mining that considers the interactions in space. It involves various techniques and methods in various areas of research. It takes into account the specificities of spatial information such as spatial relationships that can be topological, metric or directional. These relationships are implicit and difficult to represent. A Bayesian network is a graphical model that encodes causal probabilistic relationships among variables of interest, which has a powerful ability for representing and reasoning and provides an effective way to spatial data mining. Moreover, spatial data cubes allow storage and exploration of spatial data. They support spatial, non-spatial and mixed dimensions. A spatial dimension may contain vector and raster data. The spatial hierarchies can represent topological relationships between spatial objects. We propose to use Bayesian networks for knowledge discovery in spatial data cubes. The goal of our approach is first to consider spatial relationships in the data mining process, and secondly to benefit from the strength of the data warehouses to apply spatial data mining on different aggregation levels according to the topological relations between spatial data. In this article, we give a state of the art on spatial data mining and propose a framework for data mining in spatial data cubes, using Bayesian networks. We show in the proposed case study that our approach confirms the results observed in the field and it is an important way to take into account the specificities of spatial data in the spatial data mining process.