Differentiating positive and negatives stress using using brain waves

Despite the growing willingness among Singapore’s employers to offer health and productivity programmes, stress is the top health issue in the workforce today. This study focuses on identifying features from Electroencephalogram (EEG) signals to differentiate the degree of stress of a person. A multitasking framework was used as the stressor and it comprised of Stroop’s colour test, arithmetic test, and memory test. The subjects are exposed to two level of difficulty, and their EEG signals are recorded. These raw data were transformed from the time domain to the frequency domain using Fast Fourier Transform. There are 3 major bandwidth which were used in many EEG research. They are Alpha Band (7-13Hz), Beta band(14-30Hz), and Gamma Band (> 30Hz). Chebyshev Type 2 bandpass filter was used together with a Hanning window of 3 seconds to obtain these bandwidths. Stress features were added in additional of the waves, which include the band power asymmetry, average bandpower of an area, average bandpower of the left and right hemisphere of the brain. Using various modelling techniques such as random forest, support vector machine and K-nearest neighbours, we trained the model using the signal to predict the stress level of subject. 0.812 F1-score can be achieved using KNN in our experiment. This experiment also explored on how using different variables, techniques and window size during training has an impact on the accuracy of the stress level. Furthermore, my experiment identified that occipital lobe asymmetry and frontal lobe gamma has an importance role in our prediction. Popular stress features identified by previous studies were also shown to be accurate from this project. These include frontal lobe alpha asymmetry and asymmetry between hemisphere of the brain.Bachelor of Engineering (Computer Science