Fig. 5

Effects of SVM kernel type and temporal window size for a feature vector that contains all three scalp electrodes (average of all available epochs). At window sizes below 180 ms, the quadratic kernel (dashed line with square) provided maximum accuracy, and was significantly better than other kernels. At a window size of 200 ms, however, the linear classifier (solid line with circle) rose above all other classifier types. Beginning at the 700-ms temporal window, the linear SVM provided significantly better performance than any other kernel type, and reached a maximum accuracy of 94.5 % (SD = 0.064 %) when the full post-stimulus epoch (900 ms) was used. For clarity, select error bars are displayed around 180, 200, 700, and 900-ms window sizes. Error bars indicate standard deviation