12.5 AN APPLICATION OF PRINCIPAL COMPONENTS ANALYSIS
To illustrate the application of PCA, we turn to the clothing_store_PCA_training and clothing_store_PCA_test data sets. We are interested in estimating the response Sales per Visit using the predictors Purchase Visits, Days on File, Days between Purchases, Different Items Purchased, and Days since Purchase. However, Figure 12.6 shows that there is substantial correlation among the predictors. In addition, Figure 12.7, showing the regression of Sales per Visit versus the predictors, indicates some moderately inflated VIF metrics.
We therefore perform PCA on these predictors, using varimax rotation on the training data set.
Rotating the PCA solution helps in the interpretability of the components. Examining the rotated components in Figure 12.8, we find that, if...