The University of Iowa

Digital Learning Scorecard (DLSC) seeks to identify students who are struggling academically through the use of machine learning (ML) models based on student behavior metrics. Previous modeling efforts relied heavily on historical measures of academic performance, (cumulative GPA, ACT scores, etc), and omitted relevant behavioral metrics that capture both engagement and effort early in a semester. With models trained at week six in the semester, DLSC is able to identify struggling students at 78% accuracy in undergraduate courses. Used in conjunction with its interpretable insights DLSC will offer advising and student success teams a tool to help identify at-risk students and have targeted conversations with them.

DLSC uses a ‘cloud first’ modeling approach utilizing big data tools on the Google Cloud Platform (GCP) and Google’s AutoML toolkit to train its models to allow for rapid scalability. Utilizing Google’s AutoML lowers the barrier of entry to training ML models while maintaining their explainability. DLSC data source is the Unizin Data Platform, a unified institutional-level data platform that aggregates, cleans, and stores all teaching and learning data from multiple learning platforms.