The experiments have been performed using the MOA
[2] framework. The Active Classifier that implements the AL
strategies, as described in the previous section, is available
in MOA. As a base classifier we used the HoeffdingTree
that implements the popular VFDT [2]. We tested different
base learners with similar results and decided to use the
HoeffdingTree across the experiments where we control the
parameters B, the budget to label instances (e.g., 1 means
label everything, where 0.01 means 1%), and D, the class
imbalance (e.g., 0.5 means that the class is balanced, where
0.9 means 90% of the instances are noise and 10% are PD
pulses).