Moving Beyond Physiological Baselines: A New Method for Live Mental Workload Estimation

The analysis of physiological data can provide valuable information on the mental state of users interacting with a technical system, such as an intelligent tutoring system. By obtaining live estimations of mental workload a learning system can adapt, e.g., the level of difficulty of tasks to the learners needs. However, the analysis and interpretation of physiological data usually requires a baseline recording at a rested state prior to or after a task limiting their practical value. Additionally, the baseline of a physiological measure cannot be considered as a stable value but varies between days and even within a day interpersonally, so the validly calibrated data of a baseline become invalid over time limiting its value for long term use cases. This paper proposes a new method for near real time mental workload estimation. A machine learning model which predicts the mental workload based on the heart rate variability (HRV) derives metrics without the necessity of baseline recordings. First, a machine learning model is trained on a dataset of previously collected physiological data and corresponding mental workload ratings. Subsequently, physiological measures are collected continuously from a participant throughout tasks. The model is then used to predict the participant's mental workload in real time based on the HRV data. The results of our pilot study show first empirical support, that the proposed analysis technique is able to estimate mental workload in near real time with an accuracy of 90%. As this technique does not depend on baseline recordings it has the potential to be specifically valuable in applied settings such as adaptive training systems or to monitor the mental health of workers in safety-critical industries. The method could also be extrapolated for the analysis of other physiological measures in future research.