Predicting Science Achievement: A Machine Learning Approach to PISA Data Analysis Across 2006 and 2015

• In Event: AER Roundtable Session Friday 1:45 pm - Gold 4
  In Roundtable Session: Understanding Science Achievement Using Large-Scale and Longitudinal Data

Abstract

In this study, key predictors of science achievement in the PISA 2006 and 2015 assessments are investigated by focusing on school- and student-level factors. Machine Learning models (Random Forest, XGBoost, and Support Vector Regressor) are used to identify and compare the most influential predictors for student groups (low vs high-performing) across the years. SHAP values are utilized to reveal the contribution of each factor to have a better understanding of the predictors' effect. The findings emphasizes the effect of these factors over time and provide insights to address educational inequalities through the relevant policies.

Authors

• Zeynep Mentesoglu, University of Iowa

• Hyesun You, University of Iowa