Exploring Critical Predictors of Math and Science Achievement for High Achieving Students in TIMSS Data: Application of Elastic-Net Logistic Regression

In this study, we applied elastic-net logistic regression to explore critical predictors of students’ mathematics and science levels using data from eighth-grade students in the 2019 Trends in International Mathematics and Science Study (TIMSS). The elastic-net approach, a machine learning based approach combining Ridge and Least Absolute Shrinkage and Selection Operator (LASSO) regression, addresses multicollinearity while enabling variable selection, offering a robust methodology for analyzing large-scale educational datasets. Key findings highlight the critical role of home educational resources and student confidence in predicting outcomes in both subjects. Additionally, reduced disruptive behaviors during math lessons emerged as significant contributors to mathematics proficiency level. The models achieved high accuracy (up to 77.91% for mathematics and up to 70.74% in science) and balanced sensitivity and specificity, underscoring their practical relevance in educational contexts. This research demonstrates the potential of elastic-net logistic regression to provide insights into selection of predictors for student science and mathematics proficiency levels.