Using Selected Machine Learning Models to Identify and Explain Malleable Student and Teacher Antecedents of Student Achievement

• In Event: Educational Research in Transition: From Classical Methods to Machine Learning

Abstract

This study investigates the potential of machine learning (ML) methods to elucidate complex relationships between variables related to teachers and students. Rather than focusing solely on identifying the best predictive model for student math outcomes, it emphasizes understanding the underlying mechanisms of these relationships. The research posits that ML methods can uncover unique insights into what drives teacher and student learning across various contexts, providing interpretable and actionable pathways for enhancing educational theories, assessments, surveys, observation systems, teacher development plans, and the quality of evidence generated through studies. By utilizing Random Forest, XGBoost, MARS, and Shapley Values on the TEAS study data, this study reveals nuanced relationships that can contribute significantly to theory development, policy interventions, and methodological advancements.

Authors

• Amota Ataneka, University of Cincinnati

• Benjamin Kelcey, University of Cincinnati