Investigating perception dynamics and uncertainty in temporal sensory data via independent components analysis (ICA)
Wine finish from several Syrah wines are evaluated using temporal check-all-that-apply (TCATA), and results analysed with Independent Components Analysis (ICA) using the Joint Approximate Diagonalization of Eigenmatrices (JADE) algorithm.
The independent components are presented and interpreted, and suggest underlying perceptual mechanisms that are missing when the same data are analyzed via principal component analysis (PCA), suggesting that the application of ICA to sensory evaluation data may provide a more nuanced perspective. After raw and smoothed trajectories are obtained, a partial bootstrap is used to form virtual panels. Results from virtual panels are visualized to investigate uncertainty as bootstrap bands in univariate plots, and as ellipses and contrails in bivariate plots. Animated sequences facilitate review of dynamic changes over the evaluation period. Results are interpreted and compared with solutions arising from PCA, PCA with varimax rotation, PCA with oblimin transformation, and a solution from ICA with the JADE algorithm with an additional component.
Castura, J.C., Rutledge, D.N., Baker, A.K., & Ross, C.F. (2019). Investigating perception dynamics and uncertainty in temporal sensory data via independent components analysis (ICA). 13th Pangborn Sensory Science Symposium. 28 July-1 August. Edinburgh, UK. (Oral).