pose additional challenges to PD analytics. Although data
stream mining has been recently studied due to its wide-range
applications, such as sensor networks, query stream mining,
and social networks; its application on PD detection is still in
an early stage. So far, state-of-the-art PD detection algorithms
simply construct models from historical data, and then try to
apply these models to detect PD from upcoming sensed data.
These approaches only deal a single-pass challenge of data
stream mining, other challenges such as on-line training and
adaptation have yet to be properly addressed [6], [7].