Independent component analysis and regression applied on sensory data

Summary

In this paper Independent Component Analysis (ICA) and a Partial Least Squares implementation of Independent Component Regression (ICR) were applied on two sensory data sets. A brief introduction to the theory of ICA is presented, and how to find the optimal model rank is discussed. The effect of the number of independent component extracted is illustrated by comparison of ICA loadings from models with different number of components. Principal Component Analysis (PCA) and ICA were employed on the sensory data, and these methods are interpreted based on explained variance for the components and groupings of the sensory attributes. Significance testing on each sensory attribute for the components gave valuable information about the relevance in interpreting the individual attributes and components. ICA was also applied for regression purposes similar to Principal Component Regression (PCR) and Partial Least Squares Regression (PLSR). An algorithm which combines ICA and PLSR (IC-PLSR) was presented. A single response IC-PLSR seemed to be a promising complementary method to PLSR in extracting informative and valid components.