Predicting Workers' Stress: Application of a High-Performance Algorithm Using Working-Style Characteristics

Area of Science:

• Occupational Health
• Data Science
• Psychological Well-being

Background:

• Teleworking rates and work characteristics are linked to employee stress.
• Existing stress prediction models lack personalization and high performance.
• Individualized lifestyle data integration for stress modeling is underexplored.

Purpose of the Study:

• To develop a novel, high-performance algorithm for predicting employee stress.
• To create a personalized stress prediction model tailored to individual work styles.
• To evaluate the impact of teleworking rates on stress prediction accuracy.

Main Methods:

• Prospective observational study involving 190 employees over 12 weeks.
• Collected data included wearable device metrics (sleep, activity, heart rate), work shift details, and weekly questionnaires (K6, behavioral, missed meals).
• Developed a prediction model using a neighborhood cluster (N=20) based on similar working-style characteristics and teleworking rates.

Main Results:

• The proposed method achieved an Area Under the Curve (AUC) of 0.84.
• Feature importance for stress prediction varied significantly based on teleworking rates.
• High teleworking groups: stress linked to skipping lunch, work hour deviations, heart rate variability, and sleep duration.
• Low teleworking groups: stress linked to work arrival times, heart rate levels, heart rate variability, and calorie expenditure.

Conclusions:

• Clustering employees by working styles (teleworking rates) significantly improved stress prediction performance.
• The neighborhood cluster approach demonstrated validity through distinct contributing features across teleworking levels.
• Personalized stress prediction models incorporating work-style characteristics are effective.