Figure 3.

Specificities and sensitivities of different methods versus coverage. The predictions were ordered according to the increasing values of the BAGGING-STD distance to models. ASNN results were calculated using E-state indices and provided the highest balanced accuracy. J48 results were calculated using CDK descriptors and provided the highest total accuracy. LibSVM using inductive descriptors calculated the model with highest Mathew correlation coefficient. The sensitivity and specificity of models decreases with increasing values of BAGGING-STD and thus with increasing the coverage.