A Rule-Learning Algorithm of Clinical Fuzzy Production System
スポンサーリンク
概要
- 論文の詳細を見る
This paper presents a rule-learning method for fuzzy production systems performing two-group classification where each fuzzy production rule consists of an antecedent clause and a consequent clause. For the antecedent clause a histogram of the results of an experiment is usually available. A possibility distribution assessing subjective belief can be acquired from experts, but such a possibility distribution can not be considered independently of the frequency of occurrence of events. On the other hand, the construction of the membership function in the consequent clause is often more difficult, because these membership functions must be constructed from the same universal set To overcome this problem we have used a rule-learning algorithm, Examples from two groups, linguistic conditional statements of experts and fuzzy properties in the antecedent clause, are required for this rule-learning algorithm. The fuzzy properties used in this algorithm are the possibility-probability consistency proposed by Zadeh and the specificity of a fuzzy set introduced by Yager. The proposed algorithm can construct the fuzzy production rules that fit a set of examples by using fuzzy properties relative to criteria of minimum entropy and a maximum distance of sample means. Our method provides a convenient rule-learning scheme in cases where the processes of membership function tuning and formalizing decision rules are very difficult or time consuming.
- バイオメディカル・ファジィ・システム学会の論文