Model Improvement Through Transformation of Output Variables

A further improvement of the model can be achieved with a transformation (also referred to as Box-Cox transformation) of the output variables. In this case, functions such as square root, inversion, and logarithm are applied.

This allows mapping a variety of engine-based behaviors that do not behave in a linear or squared way over the input variables.

To determine the optimal transformation, all transformations are applied to the output, and the one resulting in the lowest RMSE is selected. In practice, there is already ample experience concerning the application of a specific transformation for certain output variables.