Learning Analytics for Early Identification of At-Risk Students: AMachineLearning Approach

Identifying students at risk of academic failure early enough to intervene is a central
goal of learning analytics and educational data mining (EDM). This paper reports anempirical machine learning study built on the widely used UCI Student Performancedataset (Cortez & Silva, 2008; N = 395 secondary school students). Anat-riskindicator was defined as a final grade below 10 out of 20, and three supervisedclassifiers, logistic regression, random forest, and gradient boosting, were trainedonan early-warning feature set consisting of demographic, socio-behavioral, andfirst-period academic indicators, deliberately excluding later-period grades to preservegenuine early-prediction validity. All three models achieved strong discrimination(area under the receiver operating characteristic curve, AUC, between 0.910and0.912), with gradient boosting achieving the best balance of precision and recall (F1=0.771) on a held-out test set. First-period grade, prior course failures, and absenteeismemerged as the dominant predictors of at-risk status. The paper situates these resultswithin the broader learning analytics literature on early-warning systems and discussesthe implications, limitations, and ethical considerations of deploying such models totrigger institutional support for at-risk students.