Frequency band dissociation between cognitive state recognition and person identification in EEG

Brain rhythms are used for both classifying mental states and identifying individuals. However, it is still unclear whether the same neural oscillations can successfully achieve both objectives. This study addresses this question through a three-stage experiment conducted over three weekly recording sessions. Cognitive engagement metrics first examined whether brain signals can discriminate between a cognitive task and an affective stimulus. The deep learning architectures were then trained under a leave-one-session-out scheme to evaluate which frequency bands discriminate most effectively between affective stimulation and cognitive task conditions. The same protocol assessed whether task-discriminative frequencies also preserve person-specific signatures in resting-state recordings collected one week apart. The findings indicate a strong dissociation between the two objectives. Delta-band oscillations achieved 86.9% accuracy in task classification, but only 15.6%–23.6% in person identification. On the other hand, an opposite pattern is observed in the case of broadband signals, which achieve 69.9% authentication accuracy across temporally separated sessions. Two-way ANOVA showed that the observed dissociation was not random, indicating significant band-objective interaction. Task and authentication accuracy demonstrated a significant negative correlation, with delta showing extreme task specialization and broadband demonstrating balanced versatility. The findings reveal that the engagement index differentiates task conditions with a large effect size. In contrast, the theta-alpha ratio exhibits no discriminative capacity. The results indicate that shared cognitive processes are primarily associated with slow cortical rhythms, whereas stable individual signatures extend across broader frequency bands.