Predicting U.S. Mathematics Achievement in PISA 2022: A Comparative Analysis of Penalized Regression Techniques and Identification of Key Predictors

• In Event: AI, Machine Learning, and Data Science in Modeling and Prediction: New Strategies and Applications

Abstract

This study applies machine learning techniques to examine factors influencing U.S. students’ mathematics achievement in PISA 2022. Using Ridge, LASSO, and Elastic Net regression, the analysis compares model performance and identifies key predictors across student, family, and school contexts. All models yielded comparable results, with Ridge selected for interpretability and stability. Mathematics self-efficacy, books at home, and familiarity with math concepts emerged as the strongest positive predictors. Negative predictors included school absences, grade repetition, and speaking Spanish at home. SHAP analysis illustrated individual-level predictor influence. Findings demonstrate the value of penalized regression in high-dimensional educational data and offer actionable insights for improving mathematics outcomes.

Authors

• Eter Mjavanadze, George Mason University

• Angela D. Miller, George Mason University