Understanding heterogeneity of human preferences for engineering design

In today's competitive market, it is essential for producers to provide products which not only achieve high performance, but also appeal to the tastes of consumer. Therefore, a key element of design is an understanding of human preferences for products and features. In this work, a human appraisal experiment is conducted to understand preferences for automobile occupant package design. The experiment is conducted to build predictive parametric models of consumer preferences. An issue with this class of experiment is that the heterogeneity of the experimental respondents contributes to the response, and this heterogeneity must be understood to separate the influence of design factors from that of human factors. Latent class analysis is used to combine multiple responses of the human appraisal respondents to an appropriate set of measures. Cluster analysis and smoothing spline regression are used to gain an understanding of respondent rating styles and preference heterogeneity. These analyses allow estimation of ordered logit models for prediction of consumer occupant package preferences. Methods from machine learning are also investigated as an alternative to parametric modeling.