On the Transparency of Decision-Making in Classification by Precedents With Fuzzy Descriptions
Аннотация
The article considers the problems of decision-making transparency in the presence of classification in data sets. A method for searching for and explaining hidden patterns in data is proposed, taking into account the scales of feature measurements. The values of quantitative features in the optimal boundaries of non-overlapping intervals are replaced by gradations of nominal ones. Unification of measurement scales is used to calculate the values of the membership function to classes by the frequency of occurrence of gradations of features in them. Selective deletion of feature values or introduction of gaps in the data is used to change the boundaries of intervals. The purpose of such actions is to increase confidence in the inference rules. It is proved that the presence of gaps in the data does not lead to a significant loss in the accuracy of the classification results. Three types of linguistic variables are introduced to describe confidence in the classification results, the informativeness of a set of features, and specific properties of features associated with the subject area.