I've been working through the exercises in the Applied Predictive Modeling book. The solution guide for Chapter 6 exercise 2 states that a log transformation is required to transform the response variable. https://github.com/topepo/APM_Exercises/blob/master/Ch_06.pdf
From the linked pdf:
"Before pre-processing and modeling the data, let’s examine the distribution of the permeability response (left-hand plot in Figure 4). Clearly the distribution is not symmetric, but is skewed- to-the-right. When the response has this kind of distribution, an appropriate next step is to log- transform the data. We then should verify if the transformation induced a symmetric distribution (right-hand plot in Figure 4). For these data, the log-transformation did induce an approximately symmetric distribution and we will model the log-transformed response."
Nowhere in the original textbook can I find a requirement to transform the response variable if it's not symmetric when using PLS. Is there a reference someone could share that outlines this requirement if it exists?
