Discrimination of Products: the Usefulness of Decat Function
BECHAUX Camille
DESPRES Yoann
HOLZMANN Céline
Discrimination of products: the usefulness of decat function
I Global characterization of a group of products
II Characterization of a product
III Other interesting results
Introduction:
SensoMineR is a package for sensory data analysis, witch is implemented in the R programming environment. It can be downloaded at the following addresses:
or http:// cran.r-project.org.
The decat function (DEscription of CATegories) of SensoMineR was created to identify the most characteristics variables (generally sensory descriptors) for a set of products, and for each product. It is based on variance analysis.
In the following examples, decat function used on a data frame “Orange” (12 panelists tasted 12 orange juices and rated it on sensorial attributes)
R code:
res <-decat(Orange, formul="~produit*numero_juge*seance-produit:numero_juge:seance", firstvar = 9, graph = TRUE)
I Barplots of the product effect
* A barplot of the P-values associated with the F-test of the product effect.
In this example, we can see that there is a product effect on all the descriptors, except for O.intensite: this attibute is not good to discriminate products.
*A colored table with the adjusted means of the categorical variable:
The mean of each descriptor by product is compared to the general mean, and the color shows if the mean is significantly (based on the P value):
- under the global mean (pink)
- upper the global mean (blue)
In this example, we can characterize a product:
The product number 12 is significantly characterized by high means for the descriptors "Amere", "Acide", "Intensite attaque" and significantly characterized by low means for the descriptors "Intensité.couleur", "pulpeux", "O.appeciation", "typicité.orange", "sucré", "G.appreciation". This juice can be characterized as a “strong” one, and is not really liked by panelists.
II Adjusted means
Based on this frame, we can also group products by vague category:
The orange juices 1, 11, 3 and 10 are globally characterized by a huge typicity of the orange taste, a good global appreciation, a high sweetie taste, are very pulpous and on the contrary a low bitterness and a pale color : Those are globally “sweet” juices, and they are quite liked by the panelists.
The orange juices 7, 2, 8, 6 and 9 are characterized by intermediate values for quite all the descriptors. They are intermediate juice, and they are liked by panellist but not as much as “sweet” juices.
The orange juices 5, 12 and 4 are, on the contrary of the first group, characterized by a low typicity of orange taste, a low global appreciation, a not very sweety taste, a low amount of pulpous, and on the contrary a high attack intensity, a high bitterness, acidity and a deep colour. Those are strong juices, and they the ones with the lower global appreciation.
So the conclusion of this entire frame are that panellists don’t like “hard juice”, and that they prefer sweeter one.
III Other interesting results
The decat function can also created, just by adding some r code, some graphics on others results that was created.
R code to add:
barrow(t(res$tabT), numr = 3, numc = 3)
It created a barplot of the t-statistic for a given product and a given descriptor.
For example the descriptor “Pulpeux” is good to discriminate products (high bars) whereas “O.intentsite” has little bars, the 12 products are not very different regarding this descriptor.
When a bar is in the upper half, the product has a high value for this descriptor.
The same result can be created for the coefficients resulting from the analysis of variance model for a given product and a given descriptor.
R code to add:
barrow(res$coeff, numr = 3, numc = 3, color = "orange").
Finally, the “decat” function is a good tool to characterize some products, as one product also as a member of a good of products to compare.