Generalization from uncertain and imprecise data
- Universite P. et M. Curie, Paris (France)
Most of the knowledge available about a given system is imperfect, which means imprecise, uncertain, qualitative, expressed in natural language with words which are generally vague. Some pieces of knowledge are numerical, obtained by means of measurements with more or less precise devices. They can also be incomplete, with unknown values for some elements of the system. Classification of objects, decision-making according to the description of the system, are well known problems which can be approached by various ways. Methods based on a generalization process appear very efficient when a list of already solved cases is available and sufficiently representative of all the possible cases. In this paper, we focus on the case where fuzzy sets are used to represent imperfect knowledge because of the capability of fuzzy sets to help managing imprecise data, possibly submitted to some non probabilistic uncertainty such as a subjective doubt. Fuzzy sets also present the interesting property to establish an interface between numerical and symbolic data and are interesting to use when both types of data are present. We suppose that the objects of the system are described by means of attributes, the value of which can be imprecise, uncertain or undetermined. Our purpose is to find rules enabling us to attach a class to any object of the system. We focus this study on two generalization methods based on the knowledge of a training set of objects associated with their descriptions and their classes.
- OSTI ID:
- 466437
- Report Number(s):
- CONF-9610138--
- Country of Publication:
- United States
- Language:
- English
Similar Records
Expert systems with fuzzy logic for intelligent diagnosis and control of nuclear power plants
Can fuzzy logic bring complex problems into focus? Modeling imprecise factors in environmental policy