Modeling Student Affect and Behavior using Biometric Readings, Log Files and Low Fidelity Playbacks

Document Type


Publication Date



Affective computing is computing that relates to user emotion, feelings, moods, temperament and motivation. One of its core problems that it tries to address is the automatic detection of user affect. In this paper, attempts were made to develop models of affective and behavioral states that users exhibit and experience while using Aplusix, an intelligent tutoring system for Algebra. To this end, we gathered both user interaction log data and biometrics data from first year Information Technology students at the Mapua Institute of Technology. We synchronized both logs, cut them into time frames, and labeled them following rules that we formulated for identifying the specific states of interest. We then used two supervised learning algorithms, J48 decision tree and logistic regression, to model student affect and behavior based on log files. We focused on modeling the affective states of boredom, flow and confusion, and on-task and off-task behavior. Given our data set, logistic regression resulted as the more accurate model due to better correlation as compared to J48.