Abstract
Concept formation is a machine-learning based approach to cluster analysis of symbolic data. So far, several techniques have been proposed. The system
FAVORIT, developed in our laboratory, has the ability to reflect such dynamic aspects of learning as step-by-step forgetting and strengthening the pieces of knowledge. We explain the basic features of
FAVORIT and briefly report on the experiments demonstrating the ability of our approach (1) to cope with noisy and incomplete data and (2) to simplify the internal representation of learned concepts.
