Fig. 2
From: Risk-based evaluation of machine learning-based classification methods used for medical devices
Artificial model of error distributions – graphical representations. Left side: Artificial model of error distributions, i.e. \(\:FPR\left(s\right)\) and \(\:FNR\left(s\right)\) in dependence of the threshold s. The model is based on the modified Gaussian functions as defined in Equations (1) and (2), i.e. of the form \(\:FPR\left(s\right)=\left(1-s\right)\cdot\:\text{e}\text{x}\text{p}\left(\frac{{s}^{2}}{{\sigma\:}_{FP}}\right)\) and \(\:FNR\left(s\right)=s\cdot\:\text{e}\text{x}\text{p}\left(\frac{{\left(1-s\right)}^{2}}{{\sigma\:}_{FN}}\right)\), where fixed parameters \(\:{\sigma\:}_{FN}=0.3\) were used. Right side: Resulting \(\:ROC\) curves for a set of different parameters, \(\:{\sigma\:}_{FN}=0.1, \:{\sigma\:}_{FP}=\:{\sigma\:}_{FN}=0.2,\:{\sigma\:}_{FP}=\:{\sigma\:}_{FN}=0.3\) and \(\:{\sigma\:}_{FP}=\:{\sigma\:}_{FN}=0.4\), where sFP refers to \(\:{\sigma\:}_{FP}\) and sFP to\(\:{\sigma\:}_{FN}\), in the legend