Improving Predictions of Multiple Binary Models in ILP
Figure 1
Contingency tables for a binary model (a) and a multiclass model (b), where , , and reflect a class label, actual number of examples, and predicted number of examples of the class indicated to by the subscript or the superscript, respectively. The and FN are the numbers of examples correctly and incorrectly predicted with respect to the class of interest (i.e., ). TN and FP are the numbers of examples correctly and incorrectly predicted with respect to the negative class in a binary problem only. reflects the number of classes in a multiclass problem.
