Skip to main content
Figure 9 | Journal of Cheminformatics

Figure 9

From: Using beta binomials to estimate classification uncertainty for ensemble models

Figure 9

Estimating the uncertainty profile using the averaging method. The model whose results are shown (CYP2D6-3a) has four hidden neurons, employs 28 descriptors and uses averaging to determine ensemble classification. Blue symbols show the distribution of summed network outputs for all compounds in the validation set, whereas red symbols represent network sums for misclassified compounds; tallies for the latter are offset by 0.5 for clarity. The red horizontal dotted line indicates the classification threshold that maximizes Youden’s index J. The dashed line is a linear fit of all summed network outputs to the tally of positive votes. The central box highlights tallies that straddle the classification threshold for the sum.

Back to article page