Skip to main content
. 2017 Nov 10;11:527. doi: 10.3389/fnhum.2017.00527

Table 2.

Number of samples N, after removing corrupted signals, corresponding to each electrode, and headset analyzed.

BPG BPD M32 M4S SS Total (N) Kernel σ k dominant eigenvalues k selected
Fp1 3,085 845 1,327 460 5,717 Gauss 26 5 5
Fp2 3,085 845 1,349 460 5,739 Gauss 26 5 5
F3 3,085 832 1,349 1,415 6,681 Gauss 22 4 5
F4 3,085 841 1,349 1,404 6,679 Gauss 20 4 5
C3 3,085 845 1,349 1,415 6,694 Gauss 20 5 5
C4 3,085 845 1,343 1,380 6,653 Gauss 20 5 5
P3 3,085 845 1,349 1,415 6,694 Gauss 26 4 5
P4 3,085 845 1,349 1,372 6,651 Gauss 20 5 5
Total number of PSD segments 3,085 845 1,349 460 1,415

The value σ in the Gaussian Kernel corresponds to the best performing σ for each case. All clusters were performed by selecting k = 5 cluster centers to find.

“—” Indicates the headset did not contain that electrode. Gaussian kernel: K(x,x)= exp(-||x-x||22σ2), where ||xx′||2 is the squared Euclidean distance between the two feature vectors, σ is a scaling parameter.