Learning logic functions from examples-better conceptions and models

The learning of propositional and fuzzy-logical functions and structures has been thoroughly explored in the past years and has become an efficient means for knowledge acquisition. Decision trees are broadly discussed and used, many algorithms for the learning of optimal decision trees are available. Publications and implementations, however, very often show a considerable lack of understanding of the capacities and applicability of constructed models, and one may see many applications which are developed carelessly and with little thought and come close to being dangerous mistakes. It is, however, possible to develop a methodology that is based on logical equations and makes maximum use of the existing knowledge, but avoids inadmissible generalizations and allows comprehensive knowledge engineering. The smooth transition to fuzzy-logical structures shows the efficiency of the methodology. The paper gives a comprehensive survey of the deficiencies of existing approaches and demonstrates a complete solution to all of them.