22 July 2020

Grade Prediction with Machine Learning

It has been a goal of AlmusNet to integrate key analytic data within Campus on Cloud. Functional screens for faculty and administrators are enriched with relevant KPIs. Going beyond traditional analytics, we utilize state of the art Predictive Analytics and Machine Learning to provide even more value for faculty, students and administrators.

According to Forbes, “in 2016 more than 48% of first-time, full-time students who started at a four-year college six years earlier had not yet earned a degree. For these schools, the four-year completion rate—that is, the share of students who complete a bachelor’s degree in the time the program is expected to take—is just 28%. Put another way, nearly 2 million students who begin college each year will drop out before earning a diploma. The picture at community colleges is no better. At public two-year colleges, only about 26% of full-time, first-time students complete their degree within three years” (Forbes June 6, 2018, Frederik Hess).

This presents a big problem both for students (incomplete degree programs with student debt as an added burden) and Institutions (lost revenue, fewer alumni). The same Forbes article highlights the long-standing empirical finding that students who have discussions and informal contact with faculty outside of class time are less likely to drop out of college. A system that can identify students at risk of failing or doing poorly in a course within the current term can help provide context for such informal contacts and other remedial actions to improve semester grades and prevent dropouts.

AlmusNet has developed a student grade prediction module in collaboration with partner institutions. Institutions and students that signup for this module have their student academic performance for each course processed through a Machine Learning module to predict their final course grade at mid-term.

The Machine Learning model is based on 500,000 course outcomes and will continue to grow with future versions of the model. Currently, at mid-term the model predicts the final grade at 85% accuracy for immediate grade neighbors, (for example, C- and D are immediate neighbors of D+). Where available, associated social and demographic data is included in a predicted grade report highlighting any special circumstances and overall student population trends. The predicted grade report for each student includes major reasons for the prediction.

Within Campus on Cloud, the predicted grades can be automatically displayed in the Faculty Portal, the Student Portal and the Student Mobile App.