To address the student performance problem in educational data mining, this study proposes a stacking-based RF-CatBoost model that integrates the complementary strengths of ensemble learning methods to enhance prediction accuracy and robustness. In the proposed framework, Random Forest (RF) and CatBoost are employed as the base learners to capture both global feature interactions and complex non-linear relationships within multi-source educational data. Their outputs are then stacked and fused using a combination strategy to generate the final prediction. Experimental results based on two educational datasets demonstrate that the stacking-based RF-CatBoost model consistently achieves superio... More >