Temporal Data Mining

Temporal Pattern Discovery     ◾      173

subsequences of the series [Keo05]. Discords can be used to detect anoma-

lies in an ecient way. Some interesting points are made in the article:

Discords are not necessarily found in • sparse space, which means it

would not be eective to project in an n-space and use outlier tech-

niques to nd discords.

Discords cannot be discovered by breaking the problem into smaller •

problems and recombining them to nd the solution.

Discords are a very attractive way to detect anomalies because they •

require the specication of only one parameter, the length of the

subsequence.

In the brute force algorithm for nding discords, for each possible subse-

quence the distance to the nearest non-self-match is found. ...