Session: 05-15-03: General Topics in Biomedical and Biotechnology - III

Paper Number: 95597

95597 - Engagement State Definition and Detection in Education: A Review 

Students' level of positive learning-centered affective states, like engagement or flow state, has been proved to be strongly related to dropouts' prevention, higher learning rate, and better students' performance in their courses. Human Machine Interface, health, education, among other fields, are taking significant steps towards automatically detecting the affective state of humans through physiological signals, facial expressions, and other over behaviors. Measuring users' engagement state in a more effective and user-independent way may help create a better design of interactive applications and develop intelligent, more sophisticated, and adaptative study environments. 

Engagement is often described as a multi-dimensional construct, that often involves other learning center emotions, such as boredom and frustration, but its descriptions and classification still diverge significantly in literature.  The engagement reviews found in the literature go through psychological definitions but don’t go deeper in to the physiological and behavioral indicators of the state. This review aims to analyze the current state of the art on engagement definition and detection, in order to identify which are some of the most relevant physiological and behavioral indicators for engagement in students for its prediction during presential or online courses.  

A computer-aided systematic literature search was performed following the PRISMA methodology. A total of 24 articles were selected after removing duplicates and applying the selection criteria. These studies were analyzed to extract data relative to the sensing channels that were used, the stimulus, the affective and cognitive states that were measured and how they were labeled, the number of participants, the classifier used and how accurate it was. 

The open issues in the labeling of affective states during the learning process limit the current detection and responding capabilities of the existing systems. For the engagement construct definition the correlation between its dimensions and more accurate multimodal systems is recently under development for better and innovative automatic detection methods. Even so, posture and physiological indicators, such as leaning forward or backward and parasympathetic activation (such as HR, HRV, and GSR) have proven to be strongly related to engagement and boredom states.  The multimodal channel systems have been proven to have better performance, but the question of the best channel combination is still on the table. Different classification methods (SVM, RF, NB) have achieved high accuracy performance in experimental setups, but “in the wild” studies are still a challenge for actual detection systems. Other related challenges arise with opposed enter learning emotions, such as boredom. This silent emotion has a lot of inconsistencies when measured in the literature, as valence values related to it tend to vary from study to study. 

Presenting Author: Elizabeth Rendon-Velez EAFIT

Presenting Author Biography: Elizabeth Rendón Velez is a mechanical engineer from Universidad EAFIT with a master's degree in computer engineering from the same university. Her doctorate was done at the Delft University of Technology in the Netherlands under the tutelage of Prof. Dr. Imre Horvath, on the subject of affective computing. In her professional career, she has conducted studies in the areas of computational geometry, biomedical and human factors (Man-Machine relations). Specifically, she has developed different projects on stress detection based on the measurement of the physiological response of the human body and the interaction with products. She has also participated in projects related to the design of upper and lower limb prostheses. In upper limb rehabilitation, she worked on the development of hand exoskeletons for stroke patients. In lower limb rehabilitation, she has focused on gait analysis of amputee patients when there is a relative motion between the socket and the residual limb and on the detection of abnormal gait patterns based on electromyographic signals and kinematic and spatiotemporal variables.

Authors:

Aurora Bocanumenth EAFIT
Elizabeth Rendon-Velez EAFIT