Using wearable electrodermal activity sensors and a Support Vector Machine identified non-engaged students with 81% recall, a 25 percentage point improvement over a biased random classifier.
Observational (n=24)
Wearable devices measuring electrodermal activity can reliably identify non-engaged students during lectures.
Effect estimate: 25 percentage points improvement
Absolute Event Rate: 81% vs 56%
Modern wearable devices enable the continuous and unobtrusive monitoring of human physiological parameters, including heart rate and electrodermal activity. Through the definition of adequate models these parameters allow to infer the wellbeing, empathy, or engagement of humans in different contexts. In this paper, we show that off-the-shelf wearable devices can be used to unobtrusively monitor the emotional engagement of students during lectures. We propose the use of several novel features to capture students' momentary engagement and use existing methods to characterize the general arousal of students and their physiological synchrony with the teacher. To evaluate our method we collect a data set that -- after data cleaning -- contains data from 24 students, 9 teachers, and 41 lectures. Our results show that non-engaged students can be identified with high reliability. Using a Support Vector Machine, for instance, we achieve a recall of 81% -- which is a 25 percentage points improvement with respect to a Biased Random classifier. Overall, our findings may inform the design of systems that allow students to self-monitor their engagement and act upon the obtained feedback. Teachers could profit of information about non-engaged students too to perform self-reflection and to devise and evaluate methods to (re-)engage students.
Lascio et al. (Tue,) conducted a observational in Emotional engagement (n=24). Wearable devices (electrodermal activity sensors) and Support Vector Machine vs. Biased Random classifier was evaluated on Identification of non-engaged students (recall) (25 percentage points improvement). Using wearable electrodermal activity sensors and a Support Vector Machine identified non-engaged students with 81% recall, a 25 percentage point improvement over a biased random classifier.