New concepts for fuzzy partitioning, defuzzification and derivation of probabilistic fuzzy decision trees

Abstract

Mass assignment based ID3 by Baldwin is an extension of ID3 algorithm by Quinlan for decision making and prediction problems. Mass assignment ID3 has been proved to be important while dealing with continuous variables. Use of entropy calculation to obtain better fuzzy partitions is introduced which results in asymmetric fuzzy sets. Use of asymmetric fuzzy sets, gives way to form decision trees, which increases the reliability and efficiency of the fuzzy ID3 algorithm in case of clustered databases or gives the competitive results. One attribute reduced database format is used to deal with the databases. Specific method of defuzzification is used to derive a point value from the probability distribution over the fuzzy sets of the target attribute, which becomes the prediction.