"My Brain Matters": How Multimodal Machine Learning Classifiers Reliably Predict Individual-Level but Not Class-Level Attention Measures

• In Event: Mind, Brain, and the Learning Context

Abstract

Students’ ability to pay attention to classroom content is bombarded by the presence of personal devices, overwhelming historical anxiety, and the mismatch between their typical cognitive environment, information overload, and the relatively slower pace and density of sensory information in the classroom. Our work addresses these challenges through two aims: to build predictive models of real-world classroom attention, and to understand how the experience of attention differs from its purposeful outcomes. We report a multi-modal ’in-the-wild’ study of classroom attention predicted by facial action, cardiac, and neural activity. We show that individualized datasets may be necessary to predict classroom attention, and that, surprisingly, face data may be better at predicting it than brain data collected from mobile EEG headsets.

Authors

• Joshua Friedman, Teachers College, Columbia University

• Shiang Yun Chuko, Columbia University

• Lily Penn, Columbia University

• John Thorp, Columbia University

• Helin Wang, Columbia University

• Hengbo Tong, Columbia University

• Alfredo Spagna, Columbia University