Brain computer interface based classification of stress levels
Date of Issue2016
School of Computer Engineering
Centre for High Performance Embedded Systems
This study aims to investigate whether there is any pattern for Electroencephalogram (EEG) while induced with mental stress. An automatic stress recognition system was designed and implemented with two effective stressors to induce different levels of mental stress. As a non-invasive and simple method for brain signal acquisition employed in Brain Computer Interface (BCI), EEG was used to monitor stress and the stress features were extracted from EEG signals. The Stroop colour-word test and mental arithmetic test were used as stressors to induce low level and high level of stress respectively, and their relevant C# applications were developed in Microsoft Visual Studio to interface with Emotiv Epoc device. The EEG signals from all of the 14 channels of Emotiv Epoc device were recorded during the experiment. In the meantime, appropriate data processing techniques and stress recognition algorithms were developed and implemented in Matlab to identify the different levels of stress. The power spectrum over time domain EEG signal was obtained by the Fast Fourier Transform (FFT) with Hamming window. In this study, the average of EEG power spectrum density in Theta band (4-7 Hz), Alpha band (8-12 Hz) and Beta band (13-30 Hz) were used as power features, and they were extracted using a 4 seconds sliding window with 3 seconds overlapping. Based on power features, stress features derived from bandpower asymmetry and bandpower difference is calculated and analyzed to determine the effective electrodes to recognize mental stress features. With the help of Support Vector Machine (SVM), results showed that the EEG power difference between Beta band and Alpha band produced the best average classification accuracy. The existence of stress could be identified by the change of EEG Beta and Alpha power. For 2-level stress test in Stroop colour-word test and mental arithmetic test, the average classification accuracy was 88% and 96% respectively. For 3-level stress test, it produced the best average classification accuracy of 70%. This study concluded that stress was indeed indicated by the decrease of Alpha power and increase of Beta power.
DRNTU::Engineering::Computer science and engineering::Computer applications::Social and behavioral sciences
Final Year Project (FYP)
Nanyang Technological University