A preliminary study on mental workload estimation through movement analysis using inertial sensors: A case of floor mopping

Mental workload (MWL) estimation can help prevent human error and maintain health. This study treats MWL estimation as a regression task by measuring daily MWL focused on floor cleaning in real time using behavioral information from inertial sensors. The explanatory variables include features derived from body and mopping movements, while the NASA-Task Load Index is chosen as the objective variable. The estimation pipelines are evaluated in terms of error in person-dependent model with various regression models, feature calculation windows, and sensor positions. The results show that 1) Light Gradient Boosting Machine is the best model, 2) the wrist is an optimal sensor position, and 3) the larger window size reduces the error. This study would be a starting point of MWL estimation in everyday life.