EduRiskX: A Neuro-Symbolic Framework with F-Logic Reasoning for Early Academic Risk Prediction
Abstract
Predicting academic risk in online education is essential for enabling timely interventions that improve student retention and learning outcomes. However, existing models often suffer from limited early detection capability and insufficient interpretability, resulting in a "black-box" trust crisis that hinders their adoption in real-world pedagogical settings. To address these challenges, we propose EduRiskX, a neuro-symbolic framework that integrates a temporal Transformer-based predictor with F-Logic symbolic reasoning. The neural component models longitudinal student activity sequences using temporal attention, class-weighted loss, and dynamic weekly truncation. Acting as a data-driven expert system, an F-Logic rule base—grounded in established educational theories (Engagement Theory and Student Integration Model) to mimic the diagnostic logic of human educators—is constructed exclusively from the training data. The neural risk probability and the symbolic confidence score are then combined through a logistic regression-based fusion mechanism that learns the relative contribution of each signal. Experiments on the Open University Learning Analytics Dataset (OULAD) using a strict 80/10/10 student-level split show that EduRiskX achieves an accuracy of 0.900 and an F1-score of 0.894 at the end of the semester (Week 38), with an average early detection week of 9.32 and a detection rate of 94.30%. Compared with state-of-the-art time-series models (PatchTST, iTransformer) and common deep learning baselines (LSTM, CNN), EduRiskX yields improved recall and earlier risk identification under identical conditions. Beyond predictive performance, the F-Logic module provides structured rule-based explanations linking predictions to observable behavioral patterns and educational theories.
1. Introduction
The rapid expansion of online education has created an urgent need for systems that can identify students at risk of academic failure or dropout. Early identification is critical because it enables educators to provide timely support, potentially preventing adverse outcomes. However, many existing predictive models function as "black boxes," offering little or no explanation for their decisions. This lack of transparency undermines trust and limits practical adoption by educators and administrators.
In response to these challenges, we introduce EduRiskX, a novel neuro-symbolic framework that combines the strengths of deep learning with explicit, rule-based reasoning. By integrating a temporal Transformer with F-Logic—a logic programming language designed for deductive databases and semantic web applications—EduRiskX achieves both high predictive accuracy and interpretability. The framework not only predicts whether a student is at risk but also explains why, based on observable behaviors and established educational theories.
2. Background and Related Work
Recent advances in deep learning, particularly in time-series analysis, have led to models such as PatchTST and iTransformer, which excel at capturing complex temporal patterns. However, these models are typically opaque, offering limited insight into their decision-making processes. In contrast, symbolic approaches, such as rule-based systems, are inherently interpretable but often lack the flexibility to handle large-scale, noisy educational data.
Neuro-symbolic AI seeks to bridge this gap by combining neural networks with symbolic reasoning. In the context of education, such hybrid systems can leverage the predictive power of neural models while providing transparent explanations. EduRiskX builds on this paradigm, using F-Logic to encode educational theories into a formal rule base that complements the neural predictor.
3. Methodology
3.1 Neural Component: Temporal Transformer
The neural component of EduRiskX is a temporal Transformer that processes longitudinal student activity sequences. Key features include:
- Temporal attention to focus on relevant time steps.
- Class-weighted loss to handle imbalanced data.
- Dynamic weekly truncation to adaptively consider varying sequence lengths.
This component outputs a risk probability for each student, capturing complex behavioral patterns from raw activity logs.
3.2 Symbolic Component: F-Logic Rule Base
The symbolic component is an F-Logic rule base constructed exclusively from the training data. The rules are grounded in Engagement Theory and the Student Integration Model, which describe how student engagement and integration affect persistence. These rules mimic the diagnostic reasoning of human educators, linking observable behaviors (e.g., login frequency, assignment submission, forum participation) to risk levels.
Each rule produces a confidence score, indicating the strength of the symbolic prediction.
3.3 Fusion Mechanism
The neural risk probability and symbolic confidence score are combined using a logistic regression-based fusion mechanism. This approach learns the optimal weighting of the two signals, allowing the model to balance data-driven insights with theory-based reasoning.
4. Experiments and Results
We evaluated EduRiskX on the Open University Learning Analytics Dataset (OULAD), a widely used benchmark for student success prediction. A strict 80/10/10 student-level split was employed to ensure generalization. At the end of the semester (Week 38), EduRiskX achieved:
- Accuracy: 0.900
- F1-score: 0.894
- Average early detection week: 9.32
- Detection rate: 94.30%
In comparison with state-of-the-art time-series models (PatchTST, iTransformer) and common deep learning baselines (LSTM, CNN), EduRiskX demonstrated improved recall and earlier detection under identical experimental conditions.
5. Discussion
The F-Logic component provides structured, rule-based explanations that link predictions to observable behaviors and educational theories. For example, a prediction of high risk might be explained by a rule stating that "if a student has low login frequency and missed recent assignments, then risk is high," grounded in Engagement Theory. Such explanations are crucial for building trust with educators and facilitating actionable interventions.
6. Conclusion and Future Work
EduRiskX represents a significant step toward transparent and effective academic risk prediction. By integrating neural and symbolic approaches, it achieves high performance while offering interpretable insights. Future work could explore extending the framework to other educational contexts, incorporating additional data sources, and refining the rule-learning process.
Acknowledgments
This article is an extended version of a preprint previously available on Research Square (DOI: 10.21203/rs.3.rs-8877832/v1). We thank the open-source community and the maintainers of the OULAD dataset for their contributions.
via ArXiv AI