Predicting Middle School Students' English Academic Achievement and Exploring Key Factors Using Machine Learning Techniques

Objectives This study aims to predict middle school students' academic achievement in English using various machine learning techniques and to explore the relative importance of key predictive factors by comparing and analyzing them across different grade levels. Methods To achieve this, machine learning techniques such as k-Nearest Neighbors (KNN), Decision Tree (C5.0), Random Forest (RF), Support Vector Machine (SVM), and Neural Network (NNET) were utilized. The predictive accuracy of each model was compared and analyzed for each grade level, and the key predictive factors were identified. To compare academic achievement across grade levels, vertical scale scores for each subject were used. The predictive factors were categorized into three levels: individual, family, and school. For data collection, this study used middle school data from the Gyeonggi Education Panel Study (GEPS). After data preprocessing, a total of 499 first-year, 486 second-year, and 457 third-year middle school students were included in the analysis. In total, 394 individual-level variables, 45 family-level variables, and 106 school-level variables were used to predict students' academic achievement, and the results were compared and analyzed accordingly. Results The findings indicate that Support Vector Machine (SVM) achieved the highest predictive performance for first- and second-year students, while Decision Tree (C5.0) performed best for third-year students. The analysis of key predictive factors revealed that previous-year English grades, English class comprehension, English self-efficacy, and learning attitude were consistently significant predictors across all grade levels. On the other hand, family-related factors such as parental education level and household income had relatively lower predictive power. Additionally, as grade levels increased, the importance of school environment factors (e.g., school satisfaction, classroom atmosphere) showed a gradual increase. Conclusions This study utilized various machine learning techniques to predict English academic achievement among middle school students and to compare key predictive factors across different grade levels. Furthermore, the study identified the most suitable predictive model for each grade level by comparing different machine learning algorithms. These findings provide valuable insights for developing customized learning support systems and can serve as foundational data for personalized education strategies and future education policy development.