Improving the Prediction of Undergraduate STEM Outcomes via Curricular Temporality-Based Feature Design (Poster 26)

• In Event: Friday Poster Session 9:35 am
  In Poster Session: Trends and Issues in Technology-Based Environments

Abstract

Many undergraduate students drift off their pathway toward STEM careers. In large, introductory STEM classes, instructors struggle to identify students in need of support. To address these issues, we have developed co-redesign methods to partner with disciplinary experts to create high-structure STEM courses that better support students and produce informative digital data. Applying learning analytic theory-based insights to those data, we have produced powerful features that, across multiple semesters, can identify struggling students. Here, we describe two cycles of prediction model development and application. In cycle 1, a model trained with 3 weeks of data sustained accuracy when reapplied to subsequent semesters. In cycle 2, a refit model utilizing temporal features achieved superior accuracy with just 1 week of data.

Authors

• Robert D Plumley, University of North Carolina - Chapel Hill

• Matthew L. Bernacki, University of North Carolina - Chapel Hill

• Jeff A. Greene, University of North Carolina - Chapel Hill

• Shelbi Laura Kuhlmann, University of Memphis

• Mladen Rakovic, Monash University

• Christopher J. Urban, University of Rhode Island

• Kelly Hogan, Duke University

• Chaewon Lee, University of North Carolina - Chapel Hill