2) Budget: Here we explore the impact of the budget B on
the accuracy. Figure 4 shows that with the budget of 5% the
different strategies are still able to achieve 100% accuracy at
the end of the training period except for the VarUncertainty
and FixedUncertainty. Initially the VarUncertainty strategy
outperforms all the other strategies but plateaus around 88%,
probably because it fails to learn new PD patterns. We can see
that the RandVarUncertainty does not suffer from this problem
since, as well as the other strategies that include to some extent
instances selected at random or using selective sampling