Accurate Mental Stress Detection Using Sequential Backward Selection and Adaptive Synthetic Methods.

Area of Science:

• Neuroscience
• Psychology
• Biomedical Engineering

Background:

• Mental stress significantly impacts health and work performance.
• Brain electrical activity, measured by Electroencephalogram (EEG), changes with stress.
• Resting EEG shows potential for evaluating real-life stress levels.

Purpose of the Study:

• Investigate EEG correlates of daily stress in college students.
• Evaluate the efficacy of resting EEG for stress assessment.
• Develop a robust model for stress detection using EEG data.

Main Methods:

• Longitudinal study over 13 weeks with 18 college students.
• Utilized sequential backward selection (SBS) for EEG feature selection.
• Employed adaptive synthetic (ADASYN) sampling for imbalanced stress data.

Main Results:

• Delta and theta EEG features comprised ~50% of selected features.
• Band power and pair-wise coherence (COH) achieved 94.8% accuracy in stress detection.
• ADASYN and borderline-SMOTE improved model accuracy over traditional SMOTE.

Conclusions:

• Resting EEG signals can accurately assess daily stress levels in real-world settings.
• Advanced feature selection and data balancing techniques are crucial for EEG-based stress detection.
• Findings support EEG as a tool for stress management strategies.