Mental state classification based on electroencephalogram (EEG) using multiclass support vector machine.

Introduction: Mental state refers to a person's state of mind from various perspectives, including consciousness, intention, and functionalism. Mental states closely related to everyday life include the concentrating state, neutral state, and relaxation state. Concentration is vital for cognitive tasks, while relaxation is crucial for comfort. However, individuals with mental disorders or neurological conditions often struggle to achieve these states, requiring effective detection and intervention. One method for detecting mental states is by using brainwave signals obtained through electroencephalogram (EEG). EEG has been widely used in neuroscience and clinical settings to objectively assess mental states by analyzing brainwave signals. Previous studies have demonstrated the potential of EEG-based mental state classification in stress detection, cognitive workload analysis, or depression detection.

Materials and methods: The data used in this research is secondary data in the form of recorded brainwave signals using EEG from 2018. and utilises self-reported data obtained from locally validated personal stress inventory questionnaires. The data used was obtained from four participants, including two females and two males. For preprocessing, this study uses the Hamming Windows Finite Impulse Response filtering method to extract features from each wave band. Additionally, feature selection methods are applied to choose the most relevant predictor features. Multiclass Support Vector Machine (SVM) with One-Against- One (OAO) and One-Against-All (OAA) approaches are used for classification.

Results: The feature selection process reduced the number of predictor variables from 160 to 40, focusing on minimum and maximum feature values. Multiclass SVM classification using 40 predictor variables achieved an AUC range of 0.907-0.922 (OAA) and 0.854-0.927 (OAO), while classification using all predictor variables yielded an AUC range of 0.898-0.927 (OAA) and 0.917-0.941 (OAO). Comparative performance analysis indicates that the OAA approach is superior to the OAO approach.

Conclusion: This study highlights the effectiveness of EEGbased classification of mental states using the Multiclass SVM method. The findings reinforce the role of EEG as an objective tool for mental state assessment, supporting its potential application in clinical and cognitive research for early detection of mental health disorders.