Fuzzy quantifiers in inductive learning with perception of errors in data

The authors present an extension of their earlier work (1990, 1991), in which an application of fuzzy logic with linguistic quantifiers in inductive learning from examples was proposed. That approach concerned inductive learning problems with imprecise and erroneous values of attributes and misclassifications. By using a new formulation to find a description covering, say, of almost all of the positive and almost none of the negative examples, some examples were neglected. In this paper, while including particular examples in the description sought, the authors do not take the existing values of the attributes, but further fuzzify them. In such a way, possible errors in the attribute values are accounted for by enlarging their sets of possible values. Fuzzy logic with linguistic quantifiers is used.