Using Wearable Physiological Sensors for Affect-Aware Intelligent Tutoring Systems

Abstract

Intelligent Tutoring Systems (ITS) have shown great potential in enhancing the learning process by being able to adapt to the learner's knowledge level, abilities, and difficulties. An aspect that can affect the learning process but is not taken into consideration by traditional ITS is the affective state of the learner. In this work, we propose the use of physiological signals and machine learning for the task of detecting a learner's affective state during test taking. To this end, wearable physiological sensors were used to record electroencephalography (EEG), electrocardiography (ECG), and electromyography (EMG) signals from 27 individuals while participating in a computerised English language test. Features extracted from the acquired signals were used in order to train machine learning models for the prediction of the self-reported difficulty level of the test's questions, as well as for the prediction of whether the questions would be answered correctly. Supervised classification experiments showed that there is a relation between the acquired signals and the examined tasks, reaching a classification F1-score of 74.21% for the prediction of the self-reported question difficulty level, and a classification F1-score of 59.14% for predicting whether a question was answered correctly. The acquired results demonstrate the potential of the examined approach for enhancing ITS with information relating to the affective state of the learners.