TABLE 1.

A summary of the mappings between the detected behavioral features and the state values we defined for our model.

Data Source	Feature	Observation	Metric	State Mapping
Affdex	Joy, Disgust, Contempt, Fear, Anger, Surprise, Sadness, Valence	Emotional engagement states	Max average coefficient over response period	Max average coefficients belongs to + emotion → positive
				Max average coefficients belongs to—emotion → negative
				If + and—max average emotion coefficients are equal → positive
				If (max average coefficient in low range (<0.2) AND -15 < valence < +15) → neutral
Affdex	Attention	Attentional engagement states	Average coefficient over response period	<30 → unengaged
				30–80 → semi-engaged
				>80 → fully-engaged
Kinect OR OpenPose	Angular joint velocities	Task engagement states	$\sqrt{{\displaystyle \sum _{\text{i}}\left({\text{v}}_{\text{i}}^2\right)}}$ where $i$ is the joint number	Average metric over response period >2 → high
				Average metric over response period ≤ 2
				• if the number of peaks sustained over 4 s during 50-s observation duration≥ 4 → high
				• if the number of peaks sustained over 4 s during 50-s observation duration <4 → low
				• if the number of peaks sustained over 4 s during 10-s post-prompt observation duration ≥ 1 → high
				• if the number of peaks sustained over 4 s during 10-s post-prompt observation duration <1 → low where a peak is detected when the average metric over a 4-s interval >2