Scholarlink

<p>Accurate prediction of student academic performance is essential for implementing timely interventions, yet conventional machine learning models often demonstrate inconsistent generalizability across diverse educational datasets. This study addresses this limitation by proposing a novel two-stage hybrid framework: first, a multi-classifier ensemble (decision tree, random forest, AdaBoost, or gradient boosting) generates probabilistic predictions; second, a fuzzy-logic optimization layer post-processes these predictions to explicitly handle uncertainty in student behavior and contextual factors, particularly for borderline cases. The model was rigorously evaluated on three heterogeneous student-performance datasets (<i>n</i> = 1,000–40,000) via repeated 5-fold cross-validation. The results demonstrate that the fuzzy-optimized ensemble achieves superior and consistent performance, with the highest gains on Dataset 3 (precisio: 99.46%, recal: 99.56%, F1- scoe: 99.65%). An ablation study confirms the contribution of the fuzzy layer, improving the F1- score by an averge of 1.2% across datasets. A comparative analysis with five states-of-the-art methods shows statistically significant improvements (<i>p</i> &lt; 0.05). The framework offers both predictive robustness and interpretability through transparent fuzzy rules, providing educators with actionable insights for early intervention.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Hybrid Fuzzy-Ensemble Framework for Robust Student Performance Prediction

  • 期刊论文

  • 发表日期：

    2026年3月17日

查看全文
  • Kumar Rajeswari,
  • Ponnan Vijayalakshmi

摘要

Accurate prediction of student academic performance is essential for implementing timely interventions, yet conventional machine learning models often demonstrate inconsistent generalizability across diverse educational datasets. This study addresses this limitation by proposing a novel two-stage hybrid framework: first, a multi-classifier ensemble (decision tree, random forest, AdaBoost, or gradient boosting) generates probabilistic predictions; second, a fuzzy-logic optimization layer post-processes these predictions to explicitly handle uncertainty in student behavior and contextual factors, particularly for borderline cases. The model was rigorously evaluated on three heterogeneous student-performance datasets (n = 1,000–40,000) via repeated 5-fold cross-validation. The results demonstrate that the fuzzy-optimized ensemble achieves superior and consistent performance, with the highest gains on Dataset 3 (precisio: 99.46%, recal: 99.56%, F1- scoe: 99.65%). An ablation study confirms the contribution of the fuzzy layer, improving the F1- score by an averge of 1.2% across datasets. A comparative analysis with five states-of-the-art methods shows statistically significant improvements (p < 0.05). The framework offers both predictive robustness and interpretability through transparent fuzzy rules, providing educators with actionable insights for early intervention.