Fouille de données spatiales. Approche basée sur la programmation logique inductive

Spatial data mining requires the analysis of the interactions in space. The conventional data mining algorithms do not support well this type of analysis. We present in this paper an approach based on inductive logic programming (ILP). It is based on two ideas. The first one consists in materializing these interactions using distance tables, so that the spatial data mining problem is reduced to relational data mining problem. The second consists in transforming data into first order logic, and then applying the inductive logic programming methods. This paper details this approach, and describes its application to the supervised classification by spatial decision tree. It shows also some experimentation results in the shellfish contamination analysis in Thau lagoon.