Physiological data-driven models for motion sickness prediction

Abstract

With advances in autonomous vehicle technology and in-cabin occupant monitoring systems, prediction of motion sickness (MS) has emerged as a key challenge to improve passenger experience. In this paper, a framework for MS prediction is proposed leveraging classification algorithms and timeseries physiological data, including blood volume pulse, electrodermal activity, and neck surface electromyography. The dataset used for model training contains over 1500 min of in-vehicle data, three test conditions, and a range of subject demographics. Model predictions were able to achieve 81% accuracy for binary classification (sick or not sick) and 58% for ternary classification (low, moderate or high sickness). In addition, feature importance analysis identified electrodermal activity and surface electromyography as the most relevant data streams for MS prediction. Finally, the paper analyzed the temporal dependency of physiological data on MS response and found that physiological data can precede a subject's self-reporting of MS by up to 180 s.

Keywords: Heart rate; Machine learning; Motion sickness; Surface electromyographic data.