A Deep Learning-Based Platform for Workers' Stress Detection Using Minimally Intrusive Multisensory Devices.

Area of Science:

• Human-Computer Interaction
• Occupational Health
• Artificial Intelligence

Background:

• Industry 4.0 increases human-machine interaction, creating challenges in monitoring worker stress.
• Traditional stress assessment methods (e.g., questionnaires) lack real-time capabilities.
• Existing physiological monitoring systems are often intrusive or susceptible to noise.

Purpose of the Study:

• To improve the accuracy of a minimally intrusive wearable and ambient platform for detecting worker stress.
• To leverage deep learning techniques to enhance stress level identification.
• To address the limitations of previous systems in accurately detecting multiple stress levels.

Main Methods:

• Development and refinement of a hardware-software platform for minimally intrusive physiological signal measurement.
• Implementation and comparison of three distinct neural network architectures.
• Application of a 1D-convolutional neural network (1D-CNN) for stress level classification.

Main Results:

• The 1D-CNN model achieved a high accuracy of 95.38% for identifying two distinct levels of stress.
• This represents a significant performance improvement compared to prior stress detection methods.
• The enhanced platform demonstrates improved reliability and reduced susceptibility to motion artifacts.

Conclusions:

• Deep learning, specifically 1D-CNN, significantly boosts the accuracy of non-intrusive worker stress detection.
• The improved platform offers a viable solution for real-time stress monitoring in Industry 4.0 settings.
• Accurate, continuous stress monitoring can mitigate health issues and improve worker quality of life.