EduRiskX Reports 0.900 Accuracy and Week 9.32 Early Detection on OULAD
EduRiskX fuses temporal attention with F-Logic rules on OULAD to deliver 0.900 accuracy and week-9.32 early risk detection. The neuro-symbolic design supplies rule-based explanations grounded in established educational theory. Further live-system testing is required before LMS integration.
EduRiskX combines a temporal Transformer predictor with an F-Logic rule base derived from Engagement Theory and Student Integration Model. The neural component processes weekly activity sequences under class-weighted loss and dynamic truncation. A logistic regression fusion layer weights neural probability against symbolic confidence scores generated exclusively from training data.
On the 80/10/10 student-level split of OULAD, the model exceeds PatchTST, iTransformer, LSTM, and CNN baselines in recall while advancing mean detection by multiple weeks. The F-Logic module outputs traceable rules that map behavioral sequences to educational constructs, addressing the interpretability deficit common in pure sequence models.
No production deployment records or external validation sets appear in the submission. Operational adoption requires integration with live LMS event streams and prospective calibration against institutional retention metrics beyond the 2026 arXiv preprint.
AXIOM: No peer-reviewed deployment study of EduRiskX will appear before Q4 2027.
Sources (2)
- [1]Primary Source(https://arxiv.org/abs/2608.26107)
- [2]Supporting Source(https://analyse.kmi.open.ac.uk/open_dataset)