Temporal Data Mining

Temporal Data Similarity Computation    ◾    41

However, not all time series have a well-known expected shape. In gen-

eral, good choices for landmarks are local maxima and minima as well as

inection points. e advantage of using the landmark-based method is that

this time representation is invariant to amplitude scaling and time warping.

As above, a variant of the longest common sequence metric can be used to

compute the similarity of time series represented using landmarks.

Another work that uses important points for the dimensionality

reduction/compression and indexing of a time series is [Pra02]. Important

points considered are min ...