Dina Research School

Multivariate data analysis using Principal Component Analysis

Frans van den Berg

In the field of chemometrics a popular technique for studying multivariate data tables is Principal Component Analysis. The main objective of PCA is to get a low dimensional - usually graphical - overview on large amounts of information. The bilinear PCA is a model free or soft method, meaning that no (hard) physical or theoretical relations are imposed during model building. We will study both the (essential) mathematics and the practical issues of PCA modeling, using a leading example to illustrate decisions and outcomes. The data is consumer preferences for foods collected in different countries.

If time allows we will briefly look at more ‘advanced’ methods e.g. multi-block PCA, where several data table with one mode in common are analyzed simultaneously, and multi-way PARAFAC, where data problems of three or more dimensions are analyzed.

Author: phd@dina.kvl.dk. Updated: 08 november 2001