HOW THE PARAMETER-SETTING AFFECTS THE INFERENTIAL CAPABILITY OF ARTIFICIAL NEURAL NETWORK -WITH THE CORRESPONDING RELATIONSHIP BETWEEN PRODUCTS FORM AND KANSEI IMAGES AS EXAMPLE

Abstract

ABSTRACT Artificial Neural Network (ANN) has been considered as a better model for inference in the field of Kansei Engineering research. However, the setting of parameters might strongly affect its inferential capability. For this reason, the authors researched on how the parameter-setting in ANN affected its inferential capability, and took the corresponding relationship between products form and Kansei images as example. The parameters mentioned above include three types of coding for nominal scale, four different numbers of hidden units, ten different learning rates and two opposite input and output settings. The results indicated that the coding type for nominal scale set to “on/off” type, number of hidden units set to the product of that of input units and output units, learning rate set to 1.0, formal elements (nominal scale) as input and Kansei images (rank scale) as output could achieve better analytical and inferential outcome.