Integration of different heuristics to learn concepts

The system proposed here is within the global nonsupervised induction learning field. We deal with problems defined as follows: Given a set of objects, a set of attributes, and a table of description of attribute-values of the objects; the goal is: first, to partition the set of observations in classes (clusters), and second, to discover for each cluster some general features fulfilled by its members. We successively describe the three processes of our system. First we explain the verification process. Then, the classification phase is detailed by giving the algorithm the employed and its complexity analysis. The third section describes the concept formation process based on the organisation of the clusters obtained in hierarchies. The corresponding algorithm and its worst-case complexity are also specified. The whole mechanism is illustrated throughout the text with the experimental results obtained in the contexts of chemistry and pneumonia diagnosis. Finally, we indicate future directions of research