As with conventional supervised classification, the input consists of
a set of training instances and associated class labels. In this case
however, the instances are streams. Each stream consists of a
sequence of frames. Each frame represents an instant of time of the
stream, and consists of a set of measurements or values from different
sources, termed channels. The number of frames need not be fixed.
Figure 1 illustrates the relationship between
these terms. Each stream is also labelled with its class
.
Figure: The relationship between channels, frames and
streams.
Our main goals in this work are: to produce a classifier with a low error rate, and to produce descriptions which are comprehensible to a human.