Possibilistic pattern recognition in a digestive database for mining imperfect data

We propose in this paper a method based, on the one hand, on possibility theory to calculate the similarity among the objects of any casebase, taking into account the imperfection and the heterogeneity of data, and based, on the other hand, on the geometric models like the linear and the circular unidimensional scaling and on the graphic models like the ultrametric trees in order to represent and to visualize this similarity in such a way that we can explore and discover the potential structures and patterns that exist in the data. This approach will be applied to an endoscopic casebase in order to recognize the lesions and the pathologies of this base, and several concrete examples will be given along the paper in order to clarify the mathematical concepts of the method.