Developing a Machine Learning Algorithm to Predict the Probability of Medical Staff Work Mode Using Human-Smartphone...

Area of Science:

• Human-Computer Interaction
• Machine Learning Applications
• Occupational Health

Background:

• Traditional work hour tracking methods are limited by physical presence, struggling to accurately measure remote work and breaks.
• Differentiating between on-site breaks and off-site remote work presents a significant challenge in work hour estimation.
• Machine learning offers potential for distinguishing human-smartphone interactions during work versus non-work periods.

Purpose of the Study:

• To develop and validate a novel approach, termed "probability in work mode," for precise work hour estimation.
• To leverage human-smartphone interaction patterns and GPS data for a more accurate assessment of work time.
• To differentiate between office work, remote work, and breaks using a machine learning model.

Main Methods:

• The "Staff Hours" app passively recorded participant screen events (timestamps, notifications, app usage) and GPS locations.
• Extreme gradient boosted trees and 1-dimensional convolutional neural networks were employed to process interaction patterns and generate probabilities.
• The "probability in work mode" output distinguished between office work, off-work, on-site breaks, and remote work.

Main Results:

• The study included 121 participants over 5503 person-days, achieving a model prediction performance (AUC) of 0.915.
• Work hours estimated by "probability in work mode" (11.2 hours/day) were significantly longer than GPS-defined hours (10.2 hours/day).
• The difference was primarily due to a higher estimation of remote work time (111.6 minutes) compared to break time (54.7 minutes).

Conclusions:

• The "probability in work mode" approach enhances work hour investigation accuracy by integrating human-smartphone interactions and machine learning.
• This method provides valuable insights into complex work patterns, including remote work and breaks.
• The findings suggest potential applications for optimizing work productivity and employee well-being.