EEG based brain signal analysis on stress level

• Main Author: Ng, Yu Xin
• Other Authors: Smitha Kavallur Pisharath Gopi
• Format: Final Year Project
• Language: English
• Published: 2017
• Subjects: DRNTU::Engineering::Computer science and engineering
• Online Access: http://hdl.handle.net/10356/70230
• Institution: Nanyang Technological University

In this fast pace world, one must complete multiple jobs at the same time. Stress can be easily generated when one has multiple mentally demanding jobs on hand. Experiencing stress does not always mean bad, it can be a motivating factor to succeed. However, stress overload could also cause serious health problems. Therefore, the risk of having health problem can be reduced via investigating the human brain with a stress recognition system. This study aims to investigate whether there is any pattern for Electroencephalogram (EEG) while induced with different level of stress. EEG system (Emotiv Epoc) is a very simple and effective tool use for measuring electrical activity of human brain in understanding its complex behaviour. A stress recognition system was developed with a time pressure supported multi-stressors framework to induce different level of mental stress. Emotiv is used for brain signal acquisition employed in Brain Computer Interface, EEG was used to monitor stress and extracted features out of the signal collected. Multitasking concept was used as stressors to induce different level of mental stress, where multi-stressors such Stroop colour test, mental arithmetic test and memory test were running at the same time. However, the stage, that is inducing each level of mental stress, has different question time limit. This multitasking application were developed in Microsoft Visual Studio to interface with Emotiv Epoc device. After collection of the EEG signals, data processing and positive or negative stress recognition algorithms were implemented via C# interface to Matlab to identify the level of stress occurred. In this study, the average of EED power spectrum density in Theta band (4-7Hz), Alpha band (8-12Hz) and Beta band (13-30HZ) were extracted using a 4 second sliding window with 1 second overlapping. Other stress features such as Bandpower asymmetry, bandpower difference and bandpower ratio can be derived from the power features. The classification result using the Support Vector Machine(SVM) indicated that the combination of the EEG power ratio of Beta band over Theta band and the EEG power difference between Beta band and Alpha band generated the best average classification accuracy of 77.53% in a 3-level stress classification. Comparing to the past studies, it has a higher accuracy than [2] and a lower accuracy than [1]. However, both studies used single stimuli to induce stress whereas this study used multiple stimuli to induce stress. Multiple stimuli are a closer replication of the real-world environment. This study concluded that stress can be derived by a decrease of Alpha power, an increase of Beta power and a decrease of Theta power. Even though there are a lot of studies reported in the literature, understanding stress through emotion is not discussed much. Our work is to understand stress by providing a multilevel stress platform along with the emotional aspect of the user.