A new method for classification of imprecise data using fuzzy rough fuzzification

In this paper, fuzzy rough sets are introduced as a new solution to the problem of handling imprecise input information in classification tasks. The proposed method is shown as an dispensable way to non-singleton fuzzification. Both methods are applied in neuro-fuzzy classifiers and are extended to be applied with logical-type as well as conjunction-type of fuzzy inference. Several theorems describe how to embed non-singleton fuzzification in antecedent fuzzy sets of the logical-type and conjunction-type fuzzy systems. Likewise, fuzzy rough fuzzification embeds in alike the logical-type and conjunction-type fuzzy systems. With reference to classification, the proposed neuro-fuzzy-rough structure may be considered as an extension of the neuro-rough-fuzzy structure by fuzzification of an input space, performed by fuzzy rough sets. The investigations processed for a wide range of fuzzification spread allow to observe the behavior of fuzzification methods under consideration and to verify the common certitude about the meaning of the spread.