The RV coefficients of sample and descriptor configurations increased with the number of consumers, suggesting an increase in the stability of the configurations.
However, it is recommended to check, a posteriori, the stability of sample and descriptor configurations using a stability index based on bootstrapping.
Across the 13 data sets analysed, results suggested that when working with widely different samples, 60 80 consumers can be regarded as a reasonable compromise to get stable sample and descriptor configurations.
In this context, the aim of the present work was to study the stability of sample and descriptor configurations obtained using CATA questions with consumers, by means of a bootstrapping resampling approach.
For each consumer virtual panel, sample and descriptor configurations were obtained using Correspondence Analysis and the concordance between them and the reference configurations (obtained with all the consumers) were computed using the regression vector (RV) coefficient.
However, outside of these societies, the divergence between samples and population descriptors is also problematic.
Comprehensive description of inclusion criteria and sample descriptors are needed in order to control for the potential heterogeneity in MCI aetiology.
Data from 13 consumer studies were analyzed, differing in number of consumers, product category, number of samples, degree of difference among samples and number and type of descriptors in the CATA question.
RANSAC does not feature an inherent applicability domain for individual samples although a descriptors based applicability domain approach could of course be used [31].
The definition of the domain of applicability of the training set using the vectorial representation of the training samples by molecular descriptors prevents a direct application to models trained by structured kernel-based machine learning approaches.
