Analyzing mental stress in Indian students through advanced machine learning and wearable technologies

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Machine Learning

Background:

• Mental stress is a widespread societal concern requiring accurate detection methods.
• Wearable physiological data offers a promising avenue for non-invasive stress monitoring.
• Existing methods often lack the accessibility, cost-effectiveness, or real-time capabilities needed for widespread adoption.

Purpose of the Study:

• To compare machine learning (ML) algorithms for mental stress detection using wearable physiological signals.
• To propose and validate a novel, automatic, high-performing, low-cost ML model for mental stress detection.
• To assess the performance of the developed system on benchmark datasets and in a real-time pilot study.

Main Methods:

• Collected physiological data (ECG, GSR, ST) from 200 participants under four stressors using a novel Arduino-based wearable device.
• Investigated nine ML algorithms using both multivariate and univariate features for stress classification.
• Validated the proposed model on SWELL-KW and WESAD benchmark datasets and conducted a real-time pilot test.

Main Results:

• The proposed model achieved 96.17% accuracy in detecting mental stress, with XGBoost showing superior performance in multivariate analysis.
• XGBoost demonstrated consistent accuracy in univariate analysis, highlighting its reliability for stress detection.
• Benchmark dataset validation yielded high accuracies (92.38% for SWELL-KW, 94.21% for WESAD), and a real-time pilot test achieved 97.5% accuracy with low latency.

Conclusions:

• The developed low-cost wearable device and ML model provide an effective and accessible solution for automatic mental stress detection.
• The XGBoost algorithm is a robust choice for mental stress classification using physiological signals.
• This research offers valuable insights and a reliable resource for advancing mental stress detection technologies.