Experiments

In this scientific article two different paradigms are analyzed in order to widen the scope of the transfer learning analysis. The first one relates to the Motor Imagery (MI) paradigm and the second one to the Event-Related Potential (ERP) paradigm. For the MI part 9 subjects were analyzed, each one performing two sessions, to test the accuracy of Cross-Session and Cross-Subject classification. For the ERP part, 17 subjects were analyzed and the precision of Cross-Subject classification was evaluated. In both cases, substantial improvements were obtained by introducing the procedure proposed by the article.
It should be noted that in both cases the classification methods considered are the Minimum Distance to Mean (MDM) and the Bayesian classifiers with Gaussian distribution (GM) and with mixtures of Gaussian distributions with M components (GM-M).
The use of the Cross-Session and Cross-Subject classification of MI data was experimented. To make this 9 subjects data was analyzed, resulting in a significant improvement since in some cases the performance of the Bayesian classifiers can overcome the MDM classifiers by a 30%.
In the session V-B of the article, the Cross-Subject classification results related to the P300 problem are presented.
To achieve this results 17 subjects were analyzed, where the precision index “Pr” was taken as the system evaluation variable. This index is defined by:
\(P_r=\frac{T_p}{T_p+F_p}\)
Where TP represent the “True Positives” and FP the “False Positives”.
Carrying out the tests in the article, poor results were found when the affine transformation was not performed, but when this transformation was taken into account, quite precise results were found, which can be seen in figure 1, taken from the article investigated \cite{Zanini_2017} .