Possibilistic instance-based learning

摘要

A method of instance-based learning is introduced which makes use of possibility theory and fuzzy sets. Particularly, a possibilistic version of the similarity-guided extrapolation principle underlying the instance-based learning paradigm is proposed. This version is compared to the commonly used probabilistic approach from a methodological point of view. Moreover, aspects of knowledge representation such as the modeling of uncertainty are discussed. Taking the possibilistic extrapolation principle as a point of departure, an instance-based learning procedure is outlined which includes the handling of incomplete information, methods for reducing storage requirements and the adaptation of the influence of stored cases according to their typicality. First theoretical and experimental results showing the efficiency of possibilistic instance-based learning are presented as well.