Figure 3.

Impact of scaling on classification performances for subject dependent task transfer. Two hundred four universal scaling factors between 1/1000 and 998 were used in Equation 7 to scale all the subsets in the dataset to different sizes (signal amplitudes) after trimming, and before feature extraction and classification. All models of classification were based on support vector machines with radial basis functions and 5-fold cross-validation. The dotted black vertical line pinpoints total average performances of 92.35%, and 76.33%, corresponding to Support Vector Machine (SVM) classification between High Workload vs. No Workload, based on Frontal Theta Negative and Parietal Upper Alpha Negative oscillations feature extraction, respectively, for universal scaling factor f_theta_n = f_alpha_n = 780, common to both frequency bands.