Analyzing Key Predictors of Language Anxiety: Machine Learning Models for ELAS Prediction

This study investigates the use of machine learning techniques to predict English Language Anxiety Scale (ELAS) levels among university students, motivated by the significant impact of language anxiety on educational outcomes. The research aims to enhance predictive accuracy by analyzing diverse demographic, academic, and linguistic features. The dataset from two Indonesian universities includes 292 students' educational and demographic profiles. The analysis involved three experimental phases: parameter quality assessment, feature importance evaluation, and model performance testing. Initial feature analysis identified the English Proficiency Test (EPrT), GPA, and High School Province as significant predictors of ELAS scores, with EPrT showing the highest correlation across various ELAS components. Further, eval-uating feature importance across ELAS components revealed EPrT as a consistently influential factor, highlighting its central role in language-related anxiety. In the predictive modeling phase, multiple machine learning models were tested, including Gradient Boosted Trees (GBT), Support Vector Machine (SVM), and Generalized Linear Model (GLM). GBT and SVM achieved the highest accuracy (0.843), while GLM provided a balanced trade-off between accuracy and computational efficiency. Naive Bayes, though less accurate, offered computational advantages for quick, baseline predictions. These findings underscore the relevance of EPrT and other academic indicators in predicting ELAS scores and suggest that targeted interventions based on these features could mitigate language anxiety. The study's results provide a basis for inte-grating machine learning in educational psychology, enhancing the potential for timely and personalized support for students facing language learning challenges.