From Student Risk Prediction to SC2R: Semantics-Constrained Counterfactual Recourse for Educational Decision Support
Authors: Ngoc Luyen Le, Marie-Hélène Abel, Bertrand Laforge
Organizations: Gamaizer, 93340 Le Raincy, France. · Université de Technologie de Compiègne, CNRS, Heudiasyc (Heuristics and Diagnosis of Complex Systems), CS 60319 - 60203 Compiègne Cedex, France. · Sorbonne Université, CNRS UMR 7585, LPMHE (Laboratoire de Physique Nucléaire et des Hautes Énergies), 75252 Paris cedex 05, France
Learning analytics models can identify students at risk of poor performance, but they do not directly indicate which interventions are feasible, actionable, and compatible with educational constraints. This paper introduces SC2R, a semantics-constrained counterfactual recourse framework for educational decision support. SC2R combines a calibrated predictive model, integer-programming-based recourse generation over discrete action variables, a lightweight RDF vocabulary for intervention-plan representation, and SHACL validation for enforcing timing, budget, immutability, and availability constraints. The framework is evaluated offline on the OULAD dataset using snapshots constructed relative to each assessment at two decision horizons. Results show that the predictive component provides strong performance, that compact intervention plans can be generated at scale, and that semantic validation reveals infeasible plans that lighter optimization-only settings would otherwise accept. Rather than claiming causal improvement in student outcomes, this work shows that counterfactual recourse becomes more operationally meaningful in education when recommendations are not only model-valid, but also semantically feasible and machine-checkable.
Figures & tables
Fig. 1 : Illustration of the problem formulation. Candidate plans are retained only if they satisfy both the prediction target and the semantic constraints.
Fig. 2 : Overview of SC 2 R . From student snapshots, the pipeline predicts next-assessment success, generates intervention plans, applies semantic validation, and retains only conformant plans for evaluation.
Fig. 3 : Ontology fragment for representing intervention plans, centered on cbe:ActionPlan and cbe:Action .
Intervention
Feature
Increments
Cost
Constraints
Review quizzes
Quiz clicks
+1,+2,+3
1/inc.
Before due day; available
Revisit pages
Page clicks
+1 to +5
1/2p
Before due day
Attend support
Support count
+1
3
Session available; before due day
Prep activity
Prep feature
+1
2
Schedule-compatible
TABLE I : Compact intervention vocabulary used in SC 2 R .
Item
Value
Snapshot horizons
d−14 , d−7 before next assessment
Train / val / test presentations
15/2/5
Instances per horizon (train / val / test)
190,752/29,054/84,791
Primary label
Next-assessment pass/fail ( score≥40 )
Primary recourse method
Integer programming on action variables
Full-scale IP plans scored offline
127,972
TABLE II : Dataset and protocol summary.
Fig. 4 : Predictive performance overview. Left: logistic regression performs strongly at both decision horizons, with slightly better results at d−7 . Right: supplementary nonlinear models yield only modest AUC gains on the d−14 split.
Metric
Value
Metric
Value
Plans scored
127,972
Mean number of actions
1.075
SHACL conformance rate
1.000
Mean predicted gain
0.383
Model validity rate
1.000
Mean 5-NN plausibility distance
6.950
Mean total cost
10.357
Mean noise stability
0.656
Median total cost
11.000
Mean retrain stability
0.228
TABLE III : Full-scale IP recourse summary.
Fig. 5 : Effect of semantic validation and robustness. Left: 200-case ablation comparing IP recourse, Wachter-style recourse, and IP recourse with availability constraints. Right: robustness indicators under input noise and retraining.