Explainable Machine Learning for Early Identification of Junior High School Students at Risk of Low Physics Achievement Ainun Najib, Anif Jamaluddin, and Yulianto Agung Rezeki
Master Program of Physics Education, Faculty of Teacher Training and Education, Universitas Sebelas Maret, Surakarta, Indonesia
Abstract
This study develops an explainable machine learning framework for the early identification of junior high school students at risk of low physics achievement. In this study, risk is operationally defined as a predicted physics score below the minimum mastery criterion, allowing a regression based model to function as an early warning indicator for potential learning difficulty. A preliminary dataset of 226 students was analyzed using literacy, numeracy, and socioeconomic status SES indicators as predictors of physics achievement. Four regression algorithms were compared under an 80 20 train test split: Linear Regression, Support Vector Regression SVR, Random Forest, and Gradient Boosting. Random Forest achieved the best predictive performance, with RMSE = 10.69, MAE = 8.20, and R2 = 0.32. SHAP analysis showed that literacy and numeracy were the dominant predictors, followed by parental education and occupation, indicating that both academic readiness and family background contribute to physics achievement. These findings suggest that explainable machine learning can provide not only predictive accuracy but also pedagogically actionable explanations to support early intervention in physics education.
Keywords: physics education, machine learning, SHAP, literacy, numeracy, SES
