0

I am working through An Introduction to Statistical Learning (ISLR), pdf. At "8.1.1 Regression Trees" on physical page 328, it says:

We first remove observations that are missing Salary values, and log-transform Salary so that its distribution has more of a typical bell-shape. (Recall that Salary is measured in thousands of dollars.)

It is the "log-transform" of Salary that I am asking about.

I have seen other answers on this site implying that there is no benefit in normalising the independent variables for CART regression, but could not find any information on the dependent variable.

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
MattG
  • 113
  • 4
  • 1
    For regression tree that use sum of squared errors as part of the criteria, the sum of squares has to mean something. For that the conditional distributions need to be symmetric. – Frank Harrell Sep 17 '21 at 13:15

0 Answers0