SHAP-GBM fusion model, Early warning of academic risk, Feature system construction, Interpretable Machine Learning Abstract: Early warning of academic risk has become a core methodology for enhancing educational quality due to the expansion of higher education. Current studies, however, are hampered by inadequate accuracy in conventional methodologies, limited model interpretability , challenges in synthesizing multi-dimensional data, latency issues in real-time implementation among other problems. To address these issues, this research integrates Gradient Boosting Decision Tree (LightGBM) with the SHAP interpretability framework to construct a SHAP-GBM fusion model for intelligent early-warning of academic risk. Meanwhile,based on educational evaluation theory and student development laws, the sixdimensional characteristics system and early warning mechanism, such as ideological value orientation and academic progression, are reconstructed to form a prediction-to-feedback closed loop.Experimental results on 3,681 student records show that the proposed model achieves 93.49% prediction accuracy with a Macro-F1 score of 90.87%, outperforming Random Forest, Neural Networks, and Logistic Regression by 0.55 %, 3.91 %, and 19.98 %, respectively. Moreover, statistics indicate that the recall for high-risk students reaches 85.33 % while the false alarm rate controlled at 9.34%. Additionally, the single-sample explanations require only 2.3 ms, which enables to balance the intelligent early-warning coverage and accuracy under real-time early warning requirements. Statistical significance tests confirm that SHAP-GBM is significantly better than all baselines, ensuring its reliability. https://mc.manuscriptcentral.com/jcmse Journal of Computational Methods in Science and Engineering For Peer Review Page 1 of 27 https://mc.manuscriptcentral.com/jcmse Journal of Computational Methods in Science and Engineering 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 For Peer Review 1 Research on Intelligent Early-Warning of Academic Risk Based on SHAP-GBM Fusion Model Lei Liao13, Na Long2, Qiang Hu1, Xi Yin23, Junxin Zhang2,Lianfa Zhang2,Xueqing Huang23* 1 Hunan Software Vocational and Technical University, Xiangtan 411201, China; 2 Xiangtan Institute of Technology, Xiangtan 411201, China; 3 Manuel L. Quezon University,Quezon City 1103, Philippines; * Correspondence: Xueqing Huang 1697222341@qq.com Keywords: SHAP-GBM fusion model; Early warning of academic risk; Feature system construction; Interpretable Machine Learning Abstract: Early warning of academic risk has become a core methodology for enhancing educational quality due to the expansion of higher education. Current studies, however, are hampered by inadequate accuracy in conventional methodologies, limited model interpretability , challenges in synthesizing multi-dimensional data, latency issues in real-time implementation among other problems. To address these issues, this research integrates Gradient Boosting Decision Tree (LightGBM) with the SHAP interpretability framework to construct a SHAP-GBM fusion model for intelligent early-warning of academic risk. Meanwhile,based on educational evaluation theory and student development laws, the six-dimensional characteristics system and early warning mechanism, such as ideological value orientation and academic progression, are reconstructed to form a prediction-to-feedback closed loop.Experimental results on 3,681 student records show that the proposed model achieves 93.49% prediction accuracy with a Macro-F1 score of 90.87%, outperforming Random Forest, Neural Networks, and Logistic Regression by 0.55 %, 3.91 %, and 19.98 %, respectively. Moreover, statistics indicate that the recall for high-risk students reaches 85.33 % while the false alarm rate controlled at 9.34%. Additionally, the single-sample explanations require only 2.3 ms, which enables to balance the intelligent early-warning coverage and accuracy under real-time early warning requirements. Statistical significance tests confirm that SHAP-GBM is significantly better than all baselines, ensuring its reliability. 1