RTBfoods Manual - Part 3 - Tutorial: Statistical Analyses (PCA and multiple regression) to visualise the sensory analysis data and relate it to the instrumental data. Biophysical characterization of quality traits, WP2

Bugaud Christophe, Maraval Isabelle, Meghar Karima. 2022. RTBfoods Manual - Part 3 - Tutorial: Statistical Analyses (PCA and multiple regression) to visualise the sensory analysis data and relate it to the instrumental data. Biophysical characterization of quality traits, WP2. Montpellier : RTBfoods Project-CIRAD, 24 p.

https://doi.org/10.18167/agritrop/00710

Document technique et de recherche

Prévisualisation

Version publiée - Anglais
Sous licence

.
RTBfoods_Guidance_Statistical Analyses to Visualise Sensory Data and Relate it to Instrumental Data.pdf
Télécharger (766kB) | Prévisualisation

Résumé : After sensory evaluation by a trained panel, and biophysical evaluation using the instrumental measurements of the different products, statistical treatments can be used to interpret the results. The objective of this tutorial, using XLSTAT software, is to perform and interpret 2 types of statistical treatments: (1) principal component analysis (PCA) which enables rapid visualisation of the correlations between the sensory attributes, and (2) linear regression, which allows prediction of the sensory attributes based on the biophysical (textural, biochemical) parameters. The performance of the panel has previously been checked, and the sensory data were prepared for statistical analysis (see RTBfoods_F.2.4A_Tutorial for Performance Monitoring & Sensory Data Cleaning Before Statistical Analysis_2021.pdf). The present tutorial is based on an example presented in a published Excel file that goes through one step after another. The selected PCA uses sensory data to identify major trends and sensory diversity between groups of products and between individual products. The PCA also makes it possible to measure differences between repeated products that reflect the performance of the panel (if the products are indeed identical). Multiple linear regression was used to predict sensory attributes from biophysical parameters. For this purpose, in our example, the dataset was split into two datasets: a calibration set representing ¾ of the data and a validation set containing the remaining data. Three prediction indicators were calculated to assess the accuracy and robustness of the prediction: the coefficient of determination (R²), the mean difference between observed and predicted values (RMSEC) in the calibration set, and the mean difference between the observed and predicted values in the validation set (RMSEV). The relevance of the validation and the minimum number of observations necessary to build predictive models are discussed.

Mots-clés libres : Sensory profile, Instrumental measurement, Principal component analysis, Multiple linear regression, Calibration

Agences de financement hors UE : Bill and Melinda Gates Foundation, Centre de Coopération Internationale en Recherche Agronomique pour le Développement, Centro Internacional de Agricultura Tropical, Institut National de Recherche pour l'Agriculture, l'Alimentation et l'Environnement, James Hutton Institute

Projets sur financement : (FRA) Breeding RTB Products for End User Preferences

Auteurs et affiliations