Graph-KAN : Bridging accuracy and interpretability in student performance prediction
Jitendra Kumar Guptajkgupta.cvru@gmail.comDepartment of Information Technology and Computer ScienceDr. C. V. Raman UniversityBilaspur, Chhattisgarh, 495113, IndiaView full profile → , Abhinav Shuklaamitshukla9611@gmail.comDepartment of Information Technology and Computer ScienceDr. C. V. Raman UniversityBilaspur, Chhattisgarh, 495113, IndiaView full profile → , Aryan Yadavaromsd.7781@ece.du.ac.inFaculty of TechnologyUniversity of DelhiDelhi, 110007, IndiaView full profile → , *Vanita JainCorresponding authorvanitajain@fot.du.ac.inFaculty of TechnologyUniversity of DelhiDelhi, 110007, IndiaView full profile →
* Corresponding author · click or hover a name for details
- Received:
- 01 Dec 2025
- Published Online:
- 31 Jul 2026
- Article type:
- Research Article
- Language:
- EN
- Article no.:
- JDMSC-2825
- Pages:
- 2807–2820
Abstract
The sudden growth in digital learning environments has aided the rise of student datasets that provide a great opportunity to predict and explain complex student behaviour and their performance. However, the goal of predicting academic performance remains very difficult because educational interactions are inherently very complex, multimodal, nonlinear, and imbalanced. To overcome limitations of current state-of-the-art (SOTA) models, this study proposes a Graph-KAN, a paradigm that integrates Kolmogorov- Arnold Networks (KANs) within a heterogeneous graph framework. Instead of relying on fixed activation functions, like in traditional Multi-Layer Perceptrons (MLPs), Graph-KAN takes advantage of the Kolmogorov-Arnold representation theorem to learn polynomial splines along the graph edges, allowing the model to dynamically approximate complex nonlinear education functions. We trained and validated Graph-KAN against standard baselines (GraphSAGE, GATv2, and Transformer-Conv) on the Open University Learning Analytics Dataset (OULAD). Our experimental results show that the hybrid Graph-KAN achieves a Macro F1-score of 0.536, outperforming Transformer-based architectures by about 4% and requires comparatively lower computational resources. The hybrid Graph-KAN also overcomes the limitation of class imbalance, which enhances the identification of “Distinction” students, achieving an AUC of 0.78 compared to 0.73 for Transformers. Along with high accuracy, the model offers explainability as the learned activation functions show a “diminishing returns” phenomenon in student engagement. Thus, providing educators with mathematically proven insights for personalised intervention. While basic binary prediction tasks on OULAD often yield high accuracy and F1-score, this study is based on a more complex 4-class task, which remains a challenge. This study proposes Graph-KAN as a new paradigm for such a complex setting. This work establishes KAN-based message passing as a sparingly, interpretable alternative to attention mechanisms in Educational Data Mining.
Keywords
Subject Classifications
References
[1] S. Lalwani, C. Tran, and H. Al-Rizzo, “Recent Advances in Educational Data Mining: A Comprehensive Review,” IEEE Access, vol. 12, pp. 15678–15702 (2024).
[2] J. Kuzilek, M. Hlosta, and Z. Zdrahal, “Open University Learning Analytics Dataset,” Scientific Data, vol. 4, no. 1, pp. 1–8 (2017).
[3] Y. Mouridi, M. Sadgal, and A. El Kababjie, “Predictive Modeling of Student Performance Using an Ensemble of Machine Learning Algorithms,” Computers & Education: Artificial Intelligence, vol. 4, Art. no. 100123 (2023).
[4] P. Sharma, S. K. Malik, and V. Jain, “A Comparative Analysis of Conventional Deep Learning Models with Hybrid CNN-ResNet Model for Sentiment Classification,” Journal of Discrete Mathematical Sciences & Cryptography, vol. 28, no. 6, pp. 2595–2605 (Sep. 2025).
[5] N. Malik, S. K. Malik, and V. Jain, “Semantic Web of Things for Pollution Measurement and Validation Interoperability Using AI Techniques,” Journal of Information and Optimization Sciences, vol. 45, no. 3, pp. 765–784 (Apr. 2024).
[6] S. Baatwah and A. Al-Sabaawi, “Student Performance Prediction Using Machine Learning: A Systematic Literature Review,” Education and Information Technologies, vol. 29, pp. 1–45 (2024).
[7] F. Qiu, G. Zhang, and X. Sheng, “Predicting Student Performance in Online Learning Using a Temporal-Based Attention Model,” Interactive Learning Environments, vol. 31, no. 5, pp. 2890–2905 (2023).
[8] J. Li, H. Xie, and L. Wei, “Heterogeneous Graph Neural Networks for Student Performance Prediction in Online Education,” Applied Intelligence, vol. 52, pp. 16543–16558 (2022).
[9] M. Karimi-Haghighi, C. Angulo, A. Aloraini, and R. Torné, “Predicting Student Performance in Online Learning Using Heterogeneous Graph Neural Networks,” IEEE Access, vol. 11, pp. 12345–12356 (2023).
[10] Y. Fan and T. Yamasaki, “Graph Neural Networks for Knowledge Tracing: A Comprehensive Survey,” IEEE Transactions on Learning Technologies, vol. 16, no. 4, pp. 456–472 (2023).
[11] T. K. Rusch, M. M. Bronstein, and S. Mishra, “A Survey on Oversmoothing in Graph Neural Networks,” arXiv preprint arXiv:2303.10993 (2023).
[12] M. Cheon, “Kolmogorov-Arnold Graph Neural Networks,” arXiv preprint arXiv:2406.01234 (2024).
[13] X. Zhang, Z. Liu, and H. Wang, “Complex Non-Linear Modeling of Student Learning Trajectories Using Deep Learning,” Expert Systems with Applications, vol. 238, Art. no. 121789 (2024).
[14] G. Khodabandelou and N. Aletras, “Explainable Artificial Intelligence in Education: A Comprehensive Review,” IEEE Transactions on Artificial Intelligence, vol. 4, no. 5, pp. 1123–1140 (2023).
[15] V. Swamy, M. Marras, and T. Kaser, “Trustworthy AI for Education: Inspecting the Interpretability of Learning Analytics Models,” IEEE Transactions on Learning Technologies, vol. 16, no. 5, pp. 678–690 (2023).
[16] S. A. Salloum and M. Al-Emran, “A Systematic Review of Data Balancing Techniques in Educational Data Mining,” International Journal of Information Management Data Insights, vol. 4, no. 1, Art. no. 100215 (2024).
[17] A. A. Mubarak, H. Cao, and W. Zhang, “Predictive Learning Analytics Using Deep Learning: A Review on Data Imbalance and Feature Extraction,” IEEE Access, vol. 10, pp. 23456–23470 (2022).
[18] D. Chen, Y. Lin, and W. Li, “Understanding and Mitigating Oversmoothing in Deep Graph Neural Networks,” Neural Networks, vol. 170, pp. 45–58 (2024).
[19] Z. Liu, Y. Wang, S. Vaidya, F. Ruehle, J. Halverson, M. Soljačić, T. Y. Hou, and M. Tegmark, “KAN: Kolmogorov-Arnold Networks,” arXiv preprint arXiv:2404.19756 (2024).
[20] A. Bodner, A. S. Tepsan, and J. Mathys, “Convolutional Kolmogorov-Arnold Networks,” arXiv preprint arXiv:2406.13155 (2024).
[21] R. Liu and X. Chen, “KAN-GNN: Kolmogorov-Arnold Network for Graph Neural Networks,” arXiv preprint arXiv:2405.08765 (2024).
[22] V. Krokos and R. Spliet, “On the Expressivity of Kolmogorov-Arnold Networks in Graph Learning,” arXiv preprint arXiv:2405.18765 (2024).
[23] P. Vávra and L. Pospíšil, “The Kolmogorov-Arnold Representation Theorem and Its Application to Deep Learning,” Neural Networks, vol. 175, pp. 106–118 (2024).
[24] S. S. Kusumawardani and S. A. I. Alfarozi, “Transformer Encoder Model for Sequential Prediction of Student Performance Based on Their Log Activities,” IEEE Access, vol. 11, pp. 18960–18971 (2023).
[25] M. Baranyi, M. Nagy, and R. Molontay, “Interpretable Deep Learning for Early Prediction of Student Performance,” IEEE Transactions on Learning Technologies, vol. 15, no. 3, pp. 334–346 (2022).
[26] Z. Shou, M. Xie, J. Mo, and H. Zhang, “Predicting Student Performance in Online Learning: A Multidimensional Time-Series Data Analysis Approach,” Applied Sciences, vol. 14, no. 6, Art. no. 2522 (2024).
[27] D. A. A. Silva, R. F. Santos, J. M. Oliveira, and L. P. Costa, “RITNet: Real-Time Interpretable Network for Student Performance Prediction,” IEEE Transactions on Learning Technologies, vol. 17, pp. 1–12 (2024).
