What factors enhance students' achievement? A machine learning and interpretable methods approach

Abstract

Prior research on student achievement has typically examined isolated factors or bivariate correlations, failing to capture the complex interplay between learning behaviors, pedagogical environments, and instructional design. This study addresses these limitations by employing an ensemble of five machine learning algorithms (SVM, DT, ANN, RF, and XGBoost) to model multivariate relationships between four behavioral and six instructional predictors, using final exam performance as our outcome variable. Through interpretable AI techniques, we identify several key patterns: (1) Machine learning with explainability methods effectively reveals nuanced factor-achievement relationships; (2) Behavioral metrics (hw_score, ans_score, discus_score, attend_score) show consistent positive associations; (3) High-achievers demonstrate both superior collaborative skills and preference for technology-enhanced environments; (4) Gamification frequency (s&v_num) significantly boosts outcomes; while (5) Assignment frequency (hw_num) exhibits counterproductive effects. The results advocate for: (a) teachers should balance direct instruction with active learning modalities to optimize achievement, and (b) early warning systems should leverage identifiable learning features to proactively support struggling students. Our framework enables educators to transform predictive analytics into actionable pedagogical improvements.

Fig 1. The paper’s research framework.

Fig 2. Confusion matrix of XGBoost model.

Fig 3. XGBoost-importance feature importance ranking chart.

Fig 4. The shap_value mapping plots for the top 6 features.

Fig 5. shap_value_importance figure.

Fig 6. SHAP summary_plot.

Fig 7. SHAP decision plot for samples 39-50.