Predictive assessment of student performance for early strategic guidance

In this study, we use well-accepted conceptual assessment instruments, initial state data such as the SAT, and our own recently developed instruments designed to measure aptitude in mathematics to develop a machine learning-based predictive model for student performance. Previous analysis found the expected strong correlation between performance in the mathematics and physics courses. The mathematics assessment instruments were designed to provide a means for suggesting corrective measures for students to take to improve performance in mathematics, and it was demonstrated that these measures also have an impact on performance in physics. With the predictive nature of the collected data and the impact of the various corrective measures on final grade established, we use these data to form a predictive model for student performance. By adaptively imputing missing data from previous years, and forming a random forest model, we are able to predict those students who are most at-risk of failing the introductory mathematics and physics courses with acceptable accuracy. This analysis contributes to an integrated evaluation of the current programs, which has led to an assessment-based initiative to offer strategic guidance to incoming students, better placing them for academic and career success in their selected STEM disciplines.