Improving Low-Income Students' English Proficiency in Rural China: A Predictive Achievement Model Through Machine Learning

• In Event: Opportunities to Bolster Academic Trajectories for Underserved Students

Abstract

Research identifying factors that impact students’ learning is largely based on Western theories, with methodological limitations, and rarely learning-task specific in low-resource learning contexts. Our study provides nuanced and contextualized understanding of student learning to inform local curriculum and instructional improvement policies. We construct an analytical model that uses a Machine Learning (ML) approach, multiple regression and classification algorithms, and a large dataset of 2373 student survey responses and corresponding test scores for low-income students’ English-learning achievement in rural China to identify predictive and malleable factors. We find that students’ intrinsic interest and self-efficacy are critical to determining their English learning outcomes, and that the ensemble classifier Random Forest algorithm in the binary grading system has an 83% predictive accuracy.

Authors

• Xinjie Zhang, Vanderbilt University
  Presenting Author

• Erqian Xu, University of Rochester
  Non-Presenting Author

• Xiu Cravens, Vanderbilt University
  Non-Presenting Author