Determination and classification of stress level using EEG signal and audio modalities

• Main Author: Syahrull Hi-Fi Syam, Ahmad Jamil
• Format: Thesis
• Language: English
• Published: Universiti Malaysia Perlis (UniMAP) 2014
• Subjects: Electroencephalogram (EEG); Stress Index; Eustress; k-Nearest Neighbor (k-NN)
• Online Access: http://dspace.unimap.edu.my:80/dspace/handle/123456789/33129
• Institution: Universiti Malaysia Perlis

Stress is defined as the disruption of homeostasis by physical or psychological stimuli. It can occur in two different approaches either positive way or negative way. Positive stress is called eustress and negative stress is called distress. Eustress is a positive form of stress, usually related to desirable event in person life, while distress will bring negative implication towards health on life. Thus it is essential to comprehend and come out with stress index. By knowing this, it will lead towards effective stress management and the efficiencies way of suppressing stress. This research work intends to determine the stress level (normal, very low stress, low stress, very moderate stress, moderate stress, high stress and very high stress) at 3 different sound pressure levels (60 dB, 70 dB and 80 dB) through physiological signal measurement which is Electroencephalogram signal (EEG). For stress state inducement audio clip modalities is being used. 36 sound clips which are mixed with noise selected from pilot test result, played at 3 different sound pressure levels and associated with the subjective evaluation obtained from the 30 participating subjects. EEG signal was simultaneously recorded while subjects were exposed to the played sound clips. The recorded EEG signal were analyzed and processed where features were extracted through time domain analysis (Band Energy and Approximate Entropy feature) and frequency domain analysis (Power Spectral Density feature). Theses extracted features classified through linear classifier (Linear Discriminated Analysis classifier) and non linear classifier (Neural Network and k-Nearest Neighbor classifier). The classification results by this classifier on the extracted features show the classification accuracy of the developed stress level at 3 different sound pressure levels. The classification accuracy results dwell within the range of 88.29% to 99.87%. These promising results show that the stress level were successfully developed using audio clip modalities through physiological signal measurement.