Construction of features and decision rules in fuzzy pattern recognition tasks

This paper examines the search of effective features sets that maximize an increase in the system's recognition ability. The comparison of binary rela tions composition is used as a basis for evaluation methods. Binary relatio ns compositions are obtained as a result of shadow operations on membership functions values. Special attention is paid to the study of the structure of composition graphs. The view of the graph enables one to assess the reco gnition ability of a parameter or a feature. The proposed methods enable th e evaluation of a single parameter or a group of parameters (features). Thi s allows the selection of a more precise efficient feature set and, in the long run, improves the quality of recognition.