To What Extent Has Machine Learning Achieved in Predicting At-Risk Students? Evidence From Quantitative Meta-Analysis

• In Event: Multimedia Design Considerations

Abstract

This meta-analysis aimed to provide quantitative evidence to validate to what extent machine learning techniques have been achieved in identifying online at-risk students. Additionally, meta-regressions were conducted to examine the effect of predictor data types, classical machine learning or deep learning approaches, and prediction stages as potential factors on predictive performance. Forty-two studies with 285 machine learning models were included in the meta-analysis. Overall, machine learning models were able to predict at-risk students with very good classification accuracy, with the pooled accuracy higher in the summative prediction compared to the early prediction approach. Pooled estimates of classification accuracy can be significantly enhanced in models informed by deep learning and diverse predictor data types integrating behavior, achievement, and student discussions.

Authors

• Hui Shi, Florida State University

• Secil Caskurlu, Florida State University

• Nuodi Zhang, Florida State University

• Hunhui Na, University of North Carolina - Charlotte