Predicting Students' Math Self-Efficacy: A Comparison of Traditional MLR With Machine Learning Methods

• In Event: Data Science in Education

Abstract

This study investigated factors that contributed to predicting students’ math self-efficacy by using the PISA 2012 data. Traditional multiple regression models (MLR) were compared with three regression-based machine learning methods (ridge regression, LASSO regression, and Elastic Network). A total of 4,978 students and 40 predictors were used to develop predictive models. Overall, math self-concept was identified as the most important predictor of math self-efficacy by all models. Although the machine learning algorithms did not outperform the traditional MLR in model accuracy, they took the correlations among variables into account and produced different rankings of predictor importance. In particular, more consistent results were generated between the LASSO regression and the Elastic Network.

Authors

• Ya Zhang, Western Michigan University

• Aaron Yokonia Mapondera, Western Michigan University