Methods of buildings Pareto-optimal fuzzy classifiers

The building fuzzy models process has two contradiction goals: closeness a model to the original system should be high as possible (precision of a model) and representation form of a model should be understandable (interpreting of a model). Building a fuzzy model with high precision and high interpreting at the same time is very hard process, because these properties of a model are contradictory. In the paper, suggest three criteria of type model as the fuzzy classifier. The first criterion is the index of interpreting. The index of interpreting is complex index based on geometric and positional difference between member functions of the knowledge database and linguistic divergence between fuzzy terms. The second criterion is a precision. The precision in this paper valuate as percent of right classification samples. The third criterion is a complexity (measure compactness of the knowledge database). The complexity in this paper is a count of rules in the knowledge database of the fuzzy classifier. Designed and developed algorithms helping tradeoff between these three criteria. Algorithms build many classifiers with difference ratio of interpreting index, precision and complexity. Optimal part of classifiers by each of criterion represent in the Pareto frontier. Expert can choose the one of built classifiers depending of him preference of ratio by each criterion. Result of using these algorithms getting on KEEL benchmark data sets.