This article summarizes a recently published paper. It describes how different types of errors in machine learning models can be approximated through cross-validation methods. Then, a new method of nested cross-validation is introduced, which is better suited than standard methods for estimating actually relevant error values in practice.
