Over a year ago we published an article showing how complexity can anticipate epileptic seizures. Since then our R&D has produced new complexity metrics. One such metric, designated as Cn3, has been applied recently to the same eplieptic seizure problem in the cited blog, as well as to earthquake data.
Conventional complexity anticipates seizure, as shown in the figure below.
The red dashed line indicates the onset of seizure. Complexity clearly anticipates the event by an amount of time comparable to the duration of the seizure itself. As seizure develops, complexity grows.
The new metric, Cn3 is totally different.
The signal peaks rapidly even before conventional complexity starts to rise. If we assume that complexity produces an alarm at step 250, the Cn3 metric does so between steps 100 and 150. During seizure Cn3 drops to nearly zero.
The Cn3 metrix has similar behaviour in the case of an earthquake signal as illustrated below:
In analogy to the seizure example, Cn3 peaks sharply before the quake commences while during the actual event it drops significantly.