Projection Pursuit Switching Regression for Analysis of Psychological Feelings

概要

論文の詳細を見る
Strongly nonlinear multi-variate functions, for example the artificial neural network models, hardly have insights into the local relationships between inputs and outputs. This paper proposes a weakly nonlinear version of Fuzzy c-Regression Models by Hathaway and Bezdek. Min-imization of Lagrangian function yields simultaneous estimates for the parameters of projection pursuit regression models, together with a fuzzy partitioning of mixed data. The data in each fuzzy cluster is projected in a single dimensional space and a latent weakly nonlinear (increasing, decreasing or unimodal) relationship between each independent observation and its corresponding dependent observation can be found. The proposed method is applied to an analysis of psychological feelings for T-shirt patterns.
バイオメディカル・ファジィ・システム学会の論文