The following graphs show a sample fit of data from 1 minute of EEG. Note that we compute statistics on every sample at 256 samples/second so that one minute of data produces over 15,000 data points for each parameter. The first graph is a typical fit of amplitude data to a normal distribution. It is a typical, and not a particularly great example, to be fair. The second graph is one that is somewhat better, also typical The table at the end of this article is a summary of quality of fit for 3 transformations of the data, showing that the “fourth root” gives the best fit. Each number is the correlation between the data distribution and the best Gaussian fit. Considering all channels and all components, the best fit is better than 0.99, and the worst fit is better than 0.94. This proves that even 1 minute of EEG data can provide sufficiently high quality statistics to be used for live z-score training.
