Abstract
To express decision making schemes of human beings, the authors have proposed a model based on fuzzy inference. This model fuzzily divides input spaces, which mean the evaluated values for the attributes of the objects, and the total evaluated values are realized with a linear function of the input variables in each sub-space. This paper presents a new decision making model with two more linear functions of the input variables in each sub-space. One of the two linear functions identifies the upper side of the total evaluated values and the other identifies the lower side of the total evaluation. This model can represent the variations of evaluations of human beings to the same objects. The upper side of the model corresponds to the positive evaluation and the lower does the negative one. The differences of the evaluations are represented with the coefficients of the evaluated values for the attributes. A new configuration of fuzzy neural network(FNN) which realizes this model is also presented in this paper. To identify the upper and lower sides, the modified back propagation algorithm proposed by Ishibuchi and Tanaka was used. Experiments were done using the data of impressions to facial expressions and those of the second handed motorcycles too show the feasibility of the new model. The results coincided well with the senses of the subjects. And the model could represent individual characteristics in decision making.
