
Novelty Detection in Data Streams 143
9.4.3 Approaches Based on Frequency
One way to concretize the notion of surprise in the search for unusual
patterns is to associate it to its frequency. In that sense, a pattern would be
considered surprising or anomalous if “the frequency of its occurrence differs
substantially from that expected by chance, given some previously seen data”
(Keogh et al., 2002). This definition should not be confused with that of motifs,
which are patterns that appear very frequently, i.e., are overrepresented.
Given that perspective, the TARZAN algorithm (Keogh et al., 2002) was
developed to discover surprising patterns in time