A Machine Learning–Driven Analysis of Learning Behavior and Learning Atmosphere in Predicting Junior High School English Achievement

Abstract

Accurately predicting students’ academic achievement is challenging because learning outcomes are shaped by complex interactions between individual behaviors and contextual learning environments. This study develops a machine learning framework to examine the joint effects of learning behavior and learning atmosphere on English academic performance among 256 junior high school students from three public schools in China. Survey data and standardized English scores were collected across 17 behavioral and environmental indicators. Four machine learning models—Multilayer Perceptron (MLP), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR)—were evaluated using RMSE, MAE, MAPE, and R². SVR achieved the best performance (RMSE = 1.9481, MAE = 0.7462, R² = 0.7315), outperforming the other models. SHAP analysis identified learning motivation, sleep duration, resource availability, peer influence, and parental education as the most influential predictors, while attendance, household income, and teacher quality showed relatively limited effects. These findings underscore the importance of intrinsic motivation and learning context in predicting English achievement and provide empirical support for learner-centered instruction and data-informed educational interventions under China’s “double reduction” policy.