Joint Blind Source Separation Applied to EEG Signals in the Search for Empathic Reactions
This work explores the application of Blind Source Separation (BSS) and Joint Blind Source Separation (JBSS) techniques to the preprocessing and analysis of EEG signals, focused on classification of empathic reactions. We present protocols for EEG data collection aimed at emotional studies to determine whether a person exhibits different brain electrical activity depending on whom they observe, not just the emotional situation involved. Additionally, we evaluated BSS algorithms for their effectiveness in artifact removal. In this context, we explored the application of mutual information-based BSS algorithms with kernel-based estimation. Our results indicate a significant improvement in the classification accuracy of emotional states depending on the kernel used, with Epanechnikov kernel attaining a gain of up to 10% over the classical SOBI algorithm and an absolute improvement of 21% when no artifact removal was applied. This underscores the importance of the artifact removal step in enhancing the performance of EEG signal classification. Furthermore, the analysis of JBSS methods in a simulated scenario showed the superiority of Independent Vector Analysis (IVA) based algorithms over joint diagonalization algorithms. Finally, IVA was applied on different EEG signal datasets for feature extraction, or in the context of artifact removal. Results show that IVA performance largely depends on the stepsize parameter and on the set of data being considered simultaneously. We conclude that artifact removal is crucial to improve the classification of empathic reactions in EEG, regardless of how the data is analyzed. For the classification of empathic reactions, JBSS seems inadequate due to individual variability in brain electrical activity, which compromises the exploitation of correlation between signals from different subjects.