The CID System for Concept

In this paper, we present the CID system for learning concepts from a set of examples. These examples are represented as conjunctions of attributes and their values that may be precise or not completely deened. We propose a data representation to take into account struc-tured attributes and values that may be 'impossible', 'un-known' or 'imprecise'. Lattice of fuzzy qualiiers are used to modelize this kind of information. A concept will be characterized by its extension: the set of observations that belong to the concept and its intension: the list of attribute-values related to the concept. Given a set of observations, the CID system clusters them into homogeneous groups and forms a lattice concepts. Results from Galois connexion theory are used by the clustering algorithm which is incremental. It is easy to update the set of concepts when new observations are available. Application of the CID system on sponge data from marine biology domain is presented. The CID system is implemented in C++.