My Brain Matters: How Multimodal Intra-Individual Classifiers Reliably Predict Attention, While Inter-Individual Classifiers Do Not

Poster Session F - Tuesday, April 16, 2024, 8:00 – 10:00 am EDT, Sheraton Hall ABC

Joshua Friedman¹ (jsf2167@columbia.edu), Shiang-Yun Chuko², Lily Penn³, Hafsah Shaik⁴, Conor Lee Shatto², John Thorp², Xianjue Huang⁵, Hengbo Tong⁵, Xiaofu He⁶, Alfredo Spagna²; ¹Teachers College, Columbia University, ²Department of Psychology, Columbia University, ³Cognitive Science, Columbia University, ⁴Quantitative Methods in the Social Sciences, Columbia University, ⁵Department of Statistics, Columbia University, ⁶Department of Psychiatry, Columbia University

Introduction. While the nature and impacts of attention have been hotly debated in the neuroscience community since the field’s origins, the nature of classroom attention, and how students deploy it dynamically over time to learn effectively, is still a nascent field by comparison. Method. We measured brain activity via mobile EEG headsets, facial action amplitudes, posture, and trained observer ratings in real-time in a real classroom during lecture activities to compare the predictability of objectively defined attention via in-class quiz performance and subjectively defined attention via in-class self-reports. Results. Our findings suggest that objectively defined attention is better predicted by the combination of these data than subjectively defined attention, and further, that predicting attention intra-individually (i.e., using subject A’s brain, face, body, and observer-rated data at time t to predict subject A’s attention at time t + 1 (or time t + 100) is highly reliable, while predicting attention inter-individually (i.e., using subject A’s data to predict subject B’s attention) is highly unreliable. Discussion. Our findings suggest that in-classroom models and systems designed to predict or provide feedback to teachers or students on the level of students’ attention in the classroom requires, with current technology, training classifiers on every student in the classroom, rather than a subset of them. Further, they suggest that the combination of brain, face, body, and observer data better explain variation in objective performance than subjective reports, and this carries implication for studies that exclude one or the other form of data in operationalizing classroom attention.

Topic Area: ATTENTION: Other