Optimizing Mental Stress Detection via Heart Rate Variability Feature Selection.

Area of Science:

• Physiology
• Biomedical Engineering
• Data Science

Background:

• Stress-related disorders are increasingly prevalent, requiring effective detection methods.
• Heart rate variability (HRV) analysis offers a potential non-invasive biomarker for physiological stress.
• Existing methods for stress detection may lack efficiency or accuracy.

Purpose of the Study:

• To investigate the efficacy of HRV features in detecting mental stress.
• To identify the most discriminative HRV features for stress classification.
• To develop and validate a robust stress detection model using HRV data.

Main Methods:

• Extracted 93 HRV features from electrocardiogram (ECG) signals.
• Employed filter-based, wrapper, and embedded feature selection strategies.
• Utilized Recursive Feature Elimination with Nested Leave-One-Subject-Out Cross-Validation for feature selection and model optimization.

Main Results:

• Achieved a peak F1 score of 0.76 in identifying stress states.
• Identified two key HRV features: normalized median absolute deviation of RR intervals and normalized low frequency power.
• Validated the model on an unseen dataset, yielding an average F1 score of 0.63.

Conclusions:

• Targeted feature selection is crucial for optimizing stress detection models, especially with high-dimensional HRV data.
• Specific HRV metrics demonstrate significant potential for accurate and efficient mental stress detection.
• This research contributes to developing advanced stress monitoring systems for improved mental health assessment.