Fig. 5
From: Efficiency of different measures for defining the applicability domain of classification models
Ames data set employing classification random forests. Predictiveness curves for all confidence measures and the two novelty measures \(\cos_{\alpha }\) and \(\gamma_{Euc}\) are shown. They show the dependence of the actual error rate depending on the quantile of the AD measure and can be used to set a threshold for the reject option that limits the maximum local error rate