BioVRSea was recently introduced as an unique multi-biometric system that combine Virtual Reality with a moving platform to induce Motion Sickness (MS). Electromyography (EMG) and balance features measuring the center of pressure (CoP) are among the bio-signals measured during a six segments protocol on BioVRSea. A total of 262 participants has been measured and all of them underwent an MS questionnaire to self-assess the MS relative symptoms and personal information like smoking, physical activity and Body Mass Index. From the last three data a binary lifestyle index is created and Machine Learning models are used to classify it starting from EMG and CoP groups of features taken individually and together. After an appropriate feature's selection, multiple algorithms are applied and the best results for the lifestyle index classification are reached with the K Nearest Neighbors algorithm (0.83 of maximum accuracy and 0.60 of recall) while Random Forest perform the best AUCROC (0.64). The most relevant features for the best models are the CoP ones during the second segment of the experiment, before the platform movements, and during its first light movements. These results show that an unhealthy lifestyle influences in a negative way the performance of a person in term of balance in a induced MS task. They can also be used as a preliminary input to study the influence of lifestyle in the behavior of people who suffers of serious MS problems or neurodegenerative patients using the novel BioVRSea platform.
Predicting lifestyle using BioVRSea multi-biometric paradigms
Donisi, L;
2022
Abstract
BioVRSea was recently introduced as an unique multi-biometric system that combine Virtual Reality with a moving platform to induce Motion Sickness (MS). Electromyography (EMG) and balance features measuring the center of pressure (CoP) are among the bio-signals measured during a six segments protocol on BioVRSea. A total of 262 participants has been measured and all of them underwent an MS questionnaire to self-assess the MS relative symptoms and personal information like smoking, physical activity and Body Mass Index. From the last three data a binary lifestyle index is created and Machine Learning models are used to classify it starting from EMG and CoP groups of features taken individually and together. After an appropriate feature's selection, multiple algorithms are applied and the best results for the lifestyle index classification are reached with the K Nearest Neighbors algorithm (0.83 of maximum accuracy and 0.60 of recall) while Random Forest perform the best AUCROC (0.64). The most relevant features for the best models are the CoP ones during the second segment of the experiment, before the platform movements, and during its first light movements. These results show that an unhealthy lifestyle influences in a negative way the performance of a person in term of balance in a induced MS task. They can also be used as a preliminary input to study the influence of lifestyle in the behavior of people who suffers of serious MS problems or neurodegenerative patients using the novel BioVRSea platform.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.