Detection of Student Confusion in MOOCs: Feature Extraction Methods for EEG Signal Processing

• In Event: AERA Virtual Poster Session 9
  In Poster Session: Integrating Technology in Education

Abstract

The aim of this study is to examine feature extraction methods for real-time detection of student confusion while viewing videos. The method relies on a sliding window for segmentation of EEG time series data as batches for re-scaling. We trained a probabilistic model using normalization and standardization methods for feature extraction and compared its performance to a majority class classifier as a baseline using a student-level cross-validation procedure. The best performing batch normalization procedure detected student confusion with 61.4% accuracy using a feed forward wrapper to discover the best combination of features. We discuss the implications for model deployment in the context of web-based learning environments such as MOOCs.

Authors

• Dany Aina, University of Utah

• Eric G. Poitras, Dalhousie University

• Monika Lohani, University of Utah