Chang, M., Looi, C. K., Biswas, G., Erni, J. N. (2023). Proceedings of The 1st International Conference on Metaverse and Artificial Companions in Education and Society (MetaACES 2023). Hong Kong: The Education University of Hong Kong
Abstract:
Most engagement prediction models focus on maximising prediction accuracy in a classification framework or minimising mean square error in a regression framework. However, the goal of an engagement prediction model from a user perspective is to be able to detect disengagement. In this work, we present a regression framework that encompasses the approach and evaluation metrics suitable for detecting disengagement. The disengagement level threshold and confidence level threshold are introduced to allow flexibility in the use of the tool. Moreover, the SMOGN oversampling
strategy and an undersampling strategy housed in an ensemble are proposed to tackle the imbalance data distribution typically found in engagement detection. They are evaluated on the DAiSEE dataset for benchmarking.
License type:
Publisher Copyright
Funding Info:
This research / project is supported by the A*STAR - AI3 Seed Fund
Grant Reference no. : C211118018