The purpose of this research project is to determine whether it is possible for a person’s future fatigue to be predicted given their previous rate of fatigue, within one session of working. To answer this question, we collected data from subjects performing a task in a simulated working environment. This task is intended to simulate moving boxes while working in a labor-intensive workplace, such as a warehouse. Subjects wore sensors on their wrist, torso and upper arm that tracked their movements. These sensors were used to collect data, which was then used to generate a variety of predictive models. The response, worker fatigue, was measured at regular 5-minute intervals using the Rating of Perceived Exertion (RPE). RPE captures perception of physical response to activity on a scale of 1 to 10. With the goal of determining the best model choice for predicting fatigue, a series of tree-based, regression, and classification models were tested. These models were then all compared to each other to determine which ones performed the best, and whether it was feasible to use any of these models to accurately predict the fatigue of workers. Our results show that it is possible to accurately predict worker fatigue in a simulated working environment. The best performing models were the tree-based models, which had Mean Absolute Error values with an approximate median of 1.4. This indicates that on average, the predicted RPE would deviate from actual RPE by 1 to 2 ratings. These results show that it is possible to reasonably predict worker fatigue.
Author(s): Yusuf Ozdemir, Computer Science and Data Science and Statistics Major
Advisor(s): Fadel Megahed, Information Systems and Analytics Department
You must be logged in to post a comment.