Grinding detection

[LollosoSi]

As noted in onxb#1 the current suggested electrode placement does not detect grinding, just clenching.

Lateral grinding (excursive parafunction) may be difficult to measure with this approach, as the temporalis is primarily responsible for vertical (clenching) activity.

I could verify that:

The classification output while simulating some grinding movements (moving jaw in two directions) using the current weights is closer to noise and relaxed classifications than clenching
My current classification bands are
0-50: relaxed/noise
50-80: movement, swallowing - generally used as spacing to avoid false positives
80+: clenching
Grinding is 40-60 in this scale
So the system does not detect grinding as it is now.

A couple ideas to fix this:

• Different electrode placement. Perhaps involving a more discreet and comfortable setup
• If the activity is electrically different but still measurable, we could distinguish clenching from grinding by creating another set of data and feeding it to the SVM algorithm, without moving any electrode
  -> Applying one or two chained SVM classifications could do it: one could be used to detect the muscles and/or the second can detect patterns based on output from the first one

Desirable traits of the final fix:

• Minimal invasiveness (does not increase the number of electrodes, puts existing ones in a smaller factor)
• Low costs (low effort implementation, does not require purchasing additional components)