Anxiety classification using EEG signals

• Main Author: Ong, Ching Wei
• Format: Final Year Project / Dissertation / Thesis
• Published: 2022
• Subjects: T Technology (General); TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://eprints.utar.edu.my/4828/1/fyp_ES_OCW_2022.pdf; http://eprints.utar.edu.my/4828/
• Institution: Universiti Tunku Abdul Rahman

Anxiety is a serious issue that has been affecting the mental health of people across the world. Anxiety is known as a sensation of worry, dread, and unease. It is a normal phenomenon as it happens when one starts to perspire, become agitated and anxious, and experience rapid heartbeat. However, if one is feeling anxious and stressfulcontinuously for a period of time, the anxiety may develop into an illness which is known as anxiety disorder. Usually, to diagnose anxiety disorders, doctors or psychologists would ask questions about the symptoms that one is facing. Lab tests or evaluations may also be carried out in the healthcare institution to validate the root cause of the symptoms. However, this whole process may be time consuming. Therefore, today, better alternatives to sense the symptoms of anxiety disorder continue to be learnt and discovered by the researchers.One of the recent studies to detect anxiety is through EEG acquisition. EEG which is the measure of brain signals is utilized as the input data for this project. A total of 23 subjects were involved in this study with the dataset obtained from DASPS. The scores obtained from HAM-A 1 and HAMA-2 functioned as the indicator to separate the anxiety detected into multiple classes of severity. The EEG data were filtered according to the frequency band (delta, theta, alpha, beta and gamma) and several features were extracted (band power, asymmetry and RMS) using MATLAB. These features were then applied in the anxiety classification process using all machine learning models available such as Decision Trees, Discriminant Analysis, Logistic Regression, Naïve Bayes classifiers, Support Vector Machines, Nearest Neighbour viiiclassifiers, Kernel Approximation classifiers and Ensemble classifiers. The accuracies obtained were then compared. In this project, the highest training accuracies of 73.3% and 68.4% were attained for 2-level and 4-level anxiety classification respectively by incorporating all 175 features. These features comprise 70 band power features (14 channels X 5 frequencies), 35 differential asymmetry features (7 features X 5 frequencies) and 70 mean RMS (14 channels X 5 frequencies). The test accuracy obtained for both conditions were equivalent to 50%. The test accuracy was computed by comparing the predicted results generated by each trained model with the actual anxiety classes. Hence, EEG signal research is definitely beneficial in detecting the occurrence of anxiety as the result obtained in this work is valid