View a larger image

When noisy data is subject to Correlation Dimension (d2) estimation,
spurious results may arise because noise will produce a d2 value similar to
the embedding dimension used. Thus noise will produce a d2 estimate of 3
when the data is embedded in 3 dimensions.

To check for determinism, data points are shuffled. This operation preserves
the probability distribution but generally produces a very different power
spectrum and correlation function.

Correlation Dimensional estimates are performed at increasing values of
embedding dimension. The shuffled data will show a linear relationship.
If the original data exhibits determinism, the attractor dimension will tend
to a plateau. (See Anesthesiology Web Enhancement Patient M Data versus
shuffled data).

* Sprott JC, Rowlands G. Chaos Data Analyzer. The Professional Version.
User's Manual. Raleigh NC, Physics Academic Software, 2003, 31-32