EEG spectra vs recurrence features in understanding cognitive effort

Characterizing cognitive states of a person based on surface electroencephalography (EEG) is traditionally performed by analyzing spectral power of the electrical brain activity at different frequency bands. Although frequency band activity over different electrode locations can provide valuable input for discriminating conscious and unconscious or drowsy and fully awaken brain states, it fails when fine grained cognitive state and effort estimation is required. This paper demonstrates that nonlinear features can be used to describe aspects of cognitive effort. Recurrence quantification analysis (RQA) features were used to discriminate mental relaxation, math calculations, and contemplating scientific articles. However, only when used in combination with traditional spectral features, the discrimination performance increases. The most dominant RQA features are recurrence rate and ratio, while the most dominant spectral features are relative beta and delta power.