BoxCoxTransformation
Box-Cox transformation is a family of power transformations designed to stabilize variance and make data more normally distributed. For a positive-valued variable x and a parameter λ, the transformation is Tλ(x) = (x^λ - 1)/λ if λ ≠ 0, and T0(x) = log x. Because x must be positive, data with zeros or negatives commonly require a constant offset before applying the transformation.
The transformation is often applied to the dependent variable in parametric models (for example linear regression
The value of λ is typically chosen to maximize the likelihood of the observed data under a normal
Box-Cox requires strictly positive data; with zero/negative values, a shift is needed or one can use extensions