Leveraging Statistical and Machine Learning Techniques to Uncover Determinants of English Performance in Chinese Middle School Students

Abstract

This study delves into the predictors of English proficiency among middle school students in China, utilizing the rich dataset provided by the China Education Panel Survey (CEPS). By integrating multilevel modeling and Support Vector Regression (SVR), this research scrutinizes a broad spectrum of factors at the individual, family, class, and school levels. Key predictors identified include students' native language proficiency (Chinese), cognitive aptitude, perceptions of English difficulty, attitudes toward English teachers, and parental involvement. The findings emphasize the significant impact of individual and family-level variables on English proficiency, highlighting the critical role of these modifiable factors. The comparative analysis demonstrates that while both multilevel models and SVR offer valuable insights, their combined application yields a more comprehensive understanding of the predictors and their interactions. This dual-method approach showcases the potential of integrating advanced statistical techniques with machine learning to enhance educational research, thereby informing policy and practice. The study’s insights are vital for developing targeted educational strategies and policies to support bilingual learners in China, ultimately improving language proficiency and educational outcomes.