Machine learning-based classification analysis of knowledge worker mental stress.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Human-Computer Interaction

Background:

• Workplace stress negatively impacts employee well-being and productivity.
• Objective stress measurement is challenging without disrupting work activities.
• Wearable devices offer a non-intrusive method for physiological data collection.

Purpose of the Study:

• To evaluate the performance of classifying stress and non-stress states using biosignal data from a wearable watch.
• To assess the feasibility of unobtrusive stress monitoring during work activities.
• To identify optimal features and classification methods for stress detection.

Main Methods:

• Participants wore a Galaxy Watch3 to collect heart rate (HR) and photoplethysmography data during induced stress and relaxation tasks.
• Classification models were built using k-NN, SVM, Decision Tree (DT), Logistic Regression (LR), Random Forest (RF), and Multilayer Perceptron (MLP) algorithms.
• Leave-One-Subject-Out Cross-Validation (LOSO-CV) was employed for performance evaluation.

Main Results:

• The top 9 features, including HR metrics (average, minimum), NNI, SDNN, vLF, HF, LF, LF/HF ratio, and total power, yielded the best classification performance.
• The optimal model achieved an accuracy of 0.817 and an F1 score of 0.801.
• Accurate stress state discrimination necessitates physiological data collection periods exceeding 2-3 minutes.

Conclusions:

• Wearable biosignal monitoring is a viable method for classifying stress and non-stress states in a work environment.
• Specific HR and heart rate variability features are crucial for effective stress classification.
• Adequate data sampling duration is essential for reliable stress detection using wearable technology.