Decision Making about Products Development through Consumer Preferences Modeling Based on Descriptive Characteristics of Products

Several modeling procedures are employed to explain the relationships between human taste preferences and sensory properties of products. External preference mapping (EPM) is one of them. It aims to extract high-level knowledge from two main datasets: a first hedonic data obtained from evaluations of consumers towards products and a second sensory data resulting from descriptive properties measurements collected for the same products. Classically, EPM proceeds in the following steps: first a reduction dimension method is used to extract the maximum of information contained in the sensory dataset and then construct a new perceptual space where the position of the studied product is determined according to its sensory characteristics. In a second step, a series of regressions are performed on the hedonic data to predict the preference liking on this new projection space and then understand the tendencies of consumers' preferences. Multiple Polynomial Regression (MPR) models are the traditionally used models to identify liking drivers.A primary objective of this study is to advocate the use of Generalized Linear Models (GLM), Generalized Additive Models (GAM) and Locally Weighted Scatterplot Smoothing (LOESS) that can be potentially used to identify and visually assess the relationship between hedonic and sensory data. In fact, the fitting of a simple linear or polynomial models would not be adequate in some cases and the analysis demands the greater flexibility offered by GLM, GAM and LOESS approaches.The strategy of analysis presented in this study covers a comparison between the performance of the proposed models with casual used models. Their performance have been compared using several statistical criteria in one hand, and through products positioning on EPM level lines in the other hand. The high performance of the LOESS algorithm is rather interesting and a smoothing version of EPM is proposed. A case study of cookies data is presented and demonstrates the strengths and weaknesses of applying these methods to EPM. The efficiency of the proposed method is also evaluated by a simulation study.