A Dual-Model Framework for Early Identification of At-Risk Students Using Gradient Boosting and Logistic Regression

Early identification of academically at-risk students is a critical challenge in education, as timely intervention can significantly improve student outcomes and retention. Many existing student performance prediction approaches rely on assessment data obtained late in the academic term, limiting their effectiveness for early intervention. This paper proposes a dual-model early-warning framework in which model selection is guided by the availability of assessment information. Using the UCI Student Performance dataset, a binary pass/fail prediction task is formulated and evaluated under two scenarios: a final prediction setting that includes early exam grades and a strict early-warning setting that excludes all exam-related features. Multiple machine learning models are evaluated, including Logistic Regression, Support Vector Machines, k-Nearest Neighbors, and ensemble methods. Experimental results show that Gradient Boosting achieves the strongest performance when early assessment data are available, while Logistic Regression provides superior discrimination in the strict pre-exam early-warning setting. These findings highlight the importance of aligning model choice with prediction timing and demonstrate that flexible, context-aware frameworks are more suitable than single-model approaches for practical earlywarning systems.