Sample size calculation from specified RMS of the resubstitution error for linear classifiers

Error estimation must be used to find the accuracy of a designed classifier, an issue that is critical in biomarker discovery. This paper applies analytical results derived previously by the authors concerning the exact sampling distribution of resubstitution for LDA under a Gaussianity assumption in order to investigate how the RMS for resubstitution behaves as a function of sample size. In particular, this allows one to make simple sample-size calculations based on a specified RMS.