Machine learning assessment of visually induced motion sickness levels based on multiple biosignals

Abstract Most of the existing assessments of Motion Sickness (MS), a condition caused by sensory conflicts, focus on the generation of the binary classification. However, those works are the single subject based classification results with just passable effects. In this study, 20 subjects were visually stimulated to induce motion sickness, and their subjective evaluations (mild, moderate, and severe feelings of MS) were recorded, as well as the EEG, the center of pressure(COP), and the head and waist motion trajectories, followed by extraction of features. The voting classifier utilized four types of base classifiers: 1) K-Nearest Neighbor Classifier (KNN), 2) Logistic Regression (LR), 3) Random Forest (RF) and 4) Multilayer perceptron neural network (MLPNN). The averaged accuracy and kappa of the voting classifier were 0.911 and 0.80 across 20 subjects. The multiple subjects binary classification yielded an accuracy of 0.763 and an kappa of 0.52. Multiple subjects three-levels classification refers to the classification of the degree of discomfort associated with motion sickness, with sensitivity values of 0.791/0.504/0.867 and kappa of 0.51, respectively. It is able to detect the Motion Sickness Level (MSL) without interrupting the subjects, which has certain application prospects. For the future use, it can detect the user experience over different VR devices, or the adaptive training process of relevant occupations such as drivers, pilots or astronauts.