A cross-sectional evaluation of meditation experience on electroencephalography data by artificial neural network and support vector machine classifiers

Figure 1. The experimental setting. (A) Electroencephalography signals are acquired at 6 areas through electrode leads while participant proceeds the assigned tasks. (B) The assigned tasks include 3 parts in the experiment. In “Baseline,” the ongoing activity is acquired by a counterbalanced random order of open-eyed and close-eyed movement for 4 cycles. In the “Meditation” session, the participants are asked to practice a focused breathing approach, and stay in a conscious state in the “Rest” session.

Figure 2. The demonstration of spectrum features. The spectrum features, amplitude of main peak (P), frequency of main peak (F), and mean power of alpha band (M) are demonstrated by the changed spectrums compared to the basic open-eyed condition from 1 participant.

Figure 3. The presentation of accuracy rate for different scaling criteria. The different scaling criteria are shown in circles of different colors for (A) artificial neural network and (B) support vector machine, presented with accuracy rate. Within each graph, different levels of normalization scaling are demonstrated in N10 (in light gray area), N1 (in dark gray area), and Raw (in black area). Here, 100% of accuracy rate is denoted as “Full” with a dark line.

Figure 4. The presentation of accuracy rate for variant spatial areas under N10 scaling. The spatial location of detected area is shown in circles for (A) artificial neural network and (B) support vector machine, presented with accuracy rate. Within each graph, different combinations of electroencephalography channels are demonstrated in FC combination in light gray area (referring to F3,F4,C3,C4), CO combination in dark gray area (referring to C3,C4,O1,O2), FO combination in black area (referring to F3,F4,O1,O2) and FCO combination in extra-light gray area (referring to all channels, F3, F4, C3, C4, O1, O2). Here, 100% of accuracy rate is denoted as “Full” with a dark line.

Figure 5. The presentation of accuracy rate for varying features of alpha spectrum under N10 scaling. The effect of spectrum feature is shown in circles for (A) artificial neural network and (B) support vector machine and their profiles of 2-sec sint across wsize are presented in (c) and (d) respectively, presented with accuracy rate. Within each graph, different combinations of spectrum indicator are demonstrated with mean power in alpha band (M) (in light gray area), combination of amplitude of main peak (P) and frequency of main peak (F) (in dark gray area), and combination of whole spectrum indictors (in black area). Here, 100% of accuracy rate is denoted as “Full” with thin dark line.

Figure 6. The evaluation of system stability. The histogram of accuracy rate in 100 runs is presented for artificial neural network (ANN) (in the left column) and support vector machine (SVM) (in the right column) for 3 meditation groups (S in black, J in gray, and N in white) in different scaling situations. All results are demonstrated while the data are acquired at the combination of 1.5-minute wsize and 2-second sint and are scaled in different normalized levels at N10 (A and B in the first row), N1 (C and D in the second row), and raw (E and F in the last row). X-axis denotes that the accuracy level of group prediction for each run and y-axis means the runs located at the same level of accuracy.