Fusion of expert and learnt knowledge in a framework of fuzzy labels Original Research Article

A framework for reasoning and modelling using fuzzy labels is described together with a calculus based on voting model semantics. In this framework models take the form of mass relations on joint label set space and can be inferred from data or from fuzzy label expressions. Mass relations define a probability distribution over the set of fuzzy label expressions but can also be mapped to distributions on the underlying parameter space. A method for fusing data-models with expert information in the form of both certain and uncertain knowledge is proposed and applied to test problems from the fields of data classification and reliability analysis of engineering systems.