Table 1: Overall Best Classification Rates (CR) for Actors’ Correct or Incorrect Actions and for Observers’ Observations of Correct or Incorrect Actions

Classification of Actors’ Correct or Incorrect Actions
Sub-Region SR-1
Time Interval (msec) Electrode Position Model Order (p) Overall Classification Rate - CR
MLP-ANN Classifier Using Cross-Validation FCM
-6 to 146 [1 5 7 11 20 21 30] p=4 86% 83%
-6 to 500 [1 9 10 25 30 31 32] p=4 86% 83%
-6 to 700 [7 8 11 13 16 29 32] p=4 83% 86%
Sub-Region SR-2
-6 to 146 [2 4 6 18] p=5 76% 80%
-6 to 500 [11 12] p=4 86% 83%
-6 to 700 [11 12] p=4 83% 83%
Classification of Observers’ Observations of Correct or Incorrect Actions
Sub-Region SR-1
90 to 318 [1 4 16 18] p=4 75% 78%
-6 to 500 [2 7 12 21 25 31]
[8 9 15 20 23 25]		 p=5
p=3		 80% 80%
-6 to 700 [8 11 19 23 24 30] p=5 84% 84%
Sub-Region SR-2
90 to 318 [1 4 16 18] p=4 78% 78%
-6 to 500 [2 3 4 6 14 16] p=5 84% 87%
-6 to 700 [3 5 18] p=5 84% 81%

CR results are given using the Multivariate Autoregressive/Simulated Annealing (MVAR/SA) feature extraction method for different time intervals and the best order model in conjunction with the Fuzzy C-Means (FCM) method and the Multi-layer Perceptron Artificial Neural Network (MLP-ANN) classifier using cross-validation, for both sub-regions SR-1 and SR-2. The electrodes whose AR parameters are estimated are also provided for each case.