Electro-physiological data fusion for stress detection.

In this work we describe the performance evaluation of a system for stress detection. The analysed data is acquired by following an experimental protocol designed to induce cognitive stress to the subjects. The experimental set-up included the recording of electroencephalography (EEG) and facial (corrugator and zygomatic) electromyography (EMG). In a preliminary analysis we are able to correlate EEG features (alpha asymmetry and alpha/beta ratio using only 3 channels) with the stress level of the subjects statistically (by using averages over subjects) but also on a subject-to-subject basis by using computational intelligence techniques reaching classification rates up to 79% when classifying 3 minutes takes. On a second step, we apply fusion techniques to the overall multi-modal feature set fusing the formerly mentioned EEG features with EMG energy. We show that the results improve significantly providing a more robust stress index every second. Given the achieved performance the system described in this work can be successfully applied for stress therapy when combined with virtual reality.