EEG based Classification of Long-term Stress Using Psychological Labeling

Area of Science:

• Neuroscience and Signal Processing
• Application of electroencephalography (EEG) in stress research.
• Utilizing machine learning for objective stress assessment.

Background:

• The growing need for accessible mental health solutions highlights the importance of objective stress measurement.
• Electroencephalography (EEG) offers a cost-effective and personalized approach to stress management.
• Current stress classification relies on subjective measures, necessitating more objective methods.

Purpose of the Study:

• To classify long-term stress using resting-state EEG signals and machine learning algorithms.
• To evaluate the efficacy of alpha asymmetry as a feature for stress classification.
• To compare the impact of different labeling methods (perceived stress scale vs. expert evaluation) on classification accuracy.

Main Methods:

• Acquisition of five-channel resting-state EEG recordings from stress and control groups.
• Extraction of frequency domain features, including frontal and temporal alpha and beta asymmetries.
• Application of feature selection using t-tests to identify significant discriminators.
• Classification using machine learning algorithms, with a focus on Support Vector Machines (SVM).

Main Results:

• Support Vector Machine (SVM) demonstrated high efficacy in classifying long-term stress.
• Alpha asymmetry emerged as a key feature for distinguishing between stress and control groups.
• Expert evaluation-based labeling significantly enhanced classification accuracy, reaching up to 85.20%.

Conclusions:

• Alpha asymmetry shows promise as a reliable biomarker for stress classification.
• Expert evaluation provides a more accurate basis for labeling stress in EEG studies.
• Machine learning models, particularly SVMs with alpha asymmetry, offer a robust method for objective, long-term stress detection.