The Box-Cox transformation is a family of power transformations with the purpose of making data x more normally distributed. If the smallest input value is zero or negative (which would invalidate the transform), a constant is added to all data such that the minimum input value becomes 1.
The transformation has a parameter lambda. The default value of lambda is calculated by maximizing a log likelihood function. This "optimal" value can be changed by the user, in the range -4 <= lambda <= 4.