مسار بولونيا
EN

ياسر محمد فالح محمود الخرساني

السنة عنوان البحث نشر البحث
2026 DRIFT-RESILIENT ARCHITECTURES FOR STUDENT RISK PREDICTION IN LMS BIG DATA ENVIRONMENTS: A PRACTICAL PIPELINE WITH EXPLAINABLE MODELS CyberLeninka
The operationalization of predictive analytics within Learning Management Systems (LMS) is frequently compromised by post-deployment performance decay. While high-velocity clickstream data offers granular behavioral insights, the non-stationary nature of student engagement—driven by evolving cohort dynamics, assessment permutations, and interface modifications—introduces significant concept drift. This study proposes a robust, drift-aware learning analytics pipeline designed to sustain model fidelity in production. We articulate a reference architecture that orchestrates scalable feature engineering, calibrated Gradient-Boosted Decision Trees (GBDT), and SHAP-based interpretability. Crucially, we formalize a closed-loop monitoring protocol that detects distributional shifts (covariate shift) to trigger shadow retraining and versioned rollbacks. This approach bridges the gap between static research benchmarks and dynamic, realworld educational ecosystems.