The current state of the art in on-line PD detection, still
requires high expert labour to label the electrical pulses. In
this paper, we propose a Active learning (AL) [21], [27]
approach for streaming data that aims to maintain an accurate
on-line model that is robust to class imbalances by carefully
selecting the most informative instances. A review of the active
learning literature can be found in [21] and [27] proposes how
to apply different active learning strategies in a data stream
classification setting.