Personalized Student Performance Prediction Modeling for Student Digital Twins

Intelligent Tutoring Systems (ITSs) and online education provide convenient and controlled platforms for learning, increasing accessibility, and standardizing education in and out of the classroom. For instance, many online ITSs are accessible to students 24/7, have multi-modality (desktop, mobile, virtual reality systems), and provide consistent curricula across students, classes, and schools. A growing design consideration of ITSs is the customization of the student experience. Modern systems tailor the curriculum to each student’s performance by providing remediation when students perform poorly or accelerating a student through lessons when strong proficiency is demonstrated. A challenge to these systems is that this customization is traditionally reactionary and only adjusts the student’s learning plan after the student has performed sub-optimally or overachieved. Proactive interventions often rely on a shared and generalized predictive algorithm that is trained on a collection of students’ past performances but applied to individual students. The presented work investigates the inconsistencies of these generalized predictive models to demonstrate that one algorithm does not work best for all students. This work sets the groundwork for the human digital twin, which models the individual student progression through an ITS and customizes the student performance predictive capabilities.