Real-time prediction of science student learning outcomes using machine learning classification of hemodynamics during virtual reality and online learning sessions

Current data sources used for the prediction of student outcomes average about 55% accuracy and require a significant amount of input data and time for researchers and educators to produce predictive models of student outcomes. The aim of this study is to examine how neurocognitive data collected via functional near infrared spectroscopy (fNIRS) may be used to create predictive models of student outcomes with greater speed and accuracy when using a synthetic adaptive learning environment (SALEs). Specifically, this study examines the utility of using neurocognitive data to develop student response prediction on a science content test. Participants were recruited from schools located in the United States (n = 40). Participants in the study engaged in three conditions: no content, video and virtual reality. The lesson video and virtual reality lesson provides an explanation of deoxyribonucleic acid replication. Observed neurocognitive responses were collected during each condition and used to predict the success of student responses on an assessment. On average the predicative accuracy of this approach is 85% and occur within 300 ms. Predictive error rates are less than 15%. Results of this study provides evidence to support the use of neurocognitive data for adaption of digitally presented content and how machine learning approaches and artificial intelligence may be used to classify student data in real-time as students engage with content. Results also illustrate good accuracy and capture of moment-to- moment fluctuations of cognition in real-time. These findings may help the development of artificially intelligent tutors and improve student-based learning analytics.