Learning Attention Level Prediction via Multimodal Physiological Data Using Wearable Wrist Devices

Abstract

Maintaining a high level of attention is a prerequisite of effective learning, which can significantly influence learning performance. In online learning, due to the separation of time and space between teachers and learners, it is difficult to monitor learners' attention level in a timely manner, thus leading to the reduction of education quality. Obviously, it is very important to explore automatic methods to assess learners' learning-attention level. In this study, we proposed a method to predict attention level using multimodal physiological data (i.e., blood volume pulse, inter-beat intervals, electrodermal activity and skin temperature) collected by a wearable wrist device. To achieve this purpose, 28 physiological features were extracted from multimodal physiological signals, which can reflect the activities of the human autonomic nervous system. Then, 19 features were selected by correlation analysis to form the optimal sub-feature set. Finally, seven traditional machine learning algorithms were adopted as the classifiers. The experimental results showed that SVM achieved the best accuracy with 75.86%, which was an acceptable level. This suggests that learning attention level prediction using multimodal physiological data is promising. The findings provide effective support for teachers' teaching decisions, so as to possibly improve the effect of online learning.