Predicting Student Aptitude Using Performance History

Many tutoring systems currently in use provide a wealth of information pertaining to student learning over long periods of time. Providing meaningful representations of student performance can indicate levels of knowledge and understanding that can alert instructors to potential struggling students in order to provide aid where it is needed; it is the goal of many researchers to even provide such indication preemptively in order to intervene before students become frustrated when attempting new skills. The goal of this work is to utilize student performance history to provide a means of quantizing student aptitude, defined here as the speed at which a student learns, and then using this measurement to predict the speed at which each student will learn the next skill before beginning. Observing a dataset of 21 skills, we compare two methods of predicting aptitude to majority class predictions at the skill level. Our results illustrate how our proposed methods exhibit different strengths in predicting student aptitude when compared to majority class, and may be used to direct attention to a struggling student before attempting a new skill.