COMPARATIVE ANALYSIS OF POST-HOC EXPLANATION METHODS FOR SMALL-SCALE MACHINE LEARNING MODELS IN LEARNING ANALYTICS SYSTEMS
DOI:
https://doi.org/10.32782/2412-9208-2026-2-187-208Keywords:
explainable artificial intelligence, learning analytics, educational informatics, local model explanations, model interpretability, decision trees, teacher' digital competence.Abstract
This paper addresses a practical problem of educational informatics – how to produce a transparent and reproducible explanation of a machine-learning prediction in a learning- analytics system, in a form readable by a teacher, a student, and an internal auditor. Grounded in a reference pedagogical scenario (an academic supervisor receives a ‘high- risk’ flag for a sixth-week student), the paper compares post-hoc explanation methods – LIME, LinearSHAP, TreeSHAP, KernelSHAP, Anchors, and Integrated Gradients – against the regulatory backdrop of the EU AI Act and GDPR. Evaluation criteria: computational efficiency, stability (Jaccard similarity of the top-5 features across 10 reruns), fidelity (prediction shift when top-3 features are zeroed), and plausibility. Experiments on Iris (4 features), Wine (13), Breast Cancer (30) in Python 3.12; significance – Welch’s t-test. Results: LinearSHAP runs under 0.1 ms at perfect stability and beats LIME at p = 0.045; LIME stability drops from 1.00 to 0.54 – 0.76 as dimensionality grows; KernelSHAP in high dimensions falls to 0.47 – not fit for audit; Anchors reaches the highest fidelity 0.75 at moderate stability; Integrated Gradients is deterministic but scales poorly (17 – 133 ms). None of the surveyed methods combines sub-millisecond time, perfect stability, and an IF-THEN format simultaneously – this gap is closed by the authors’ method Greedy-Prune-Explain (GPE): a three-phase algorithm with O(d²·n) complexity, a precision ≥ τ guarantee, and deterministic output. Expected pedagogical impact – shorter time-to- intervention, support for self-regulated learning, and documentary-grade reproducibility of pedagogical decisions. The paper concludes with practical guidelines for designers of educational information systems.
References
Aas K., Jullum M., Løland A. Explaining individual predictions when features are dependent: More accurate approximations to Shapley values. Artificial Intelligence. 2021. Vol. 298. Art. 103502. DOI: https://doi.org/10.1016/j.artint.2021.103502
Adadi A., Berrada M. Peeking Inside the Black-Box: A Survey on Explainable Artificial Intelligence (XAI). IEEE Access. 2018. Vol. 6. P. 52138–52160. DOI: https://doi.org/10.1109/ACCESS.2018.2870052
Adebayo J., Gilmer J., Muelly M., Goodfellow I., Hardt M., Kim B. Sanity Checks for Saliency Maps. Proceedings of the 32nd Conference on Neural Information Processing Systems. Montréal, Canada, 2018.
Alvarez-Melis D., Jaakkola T. S. On the Robustness of Interpretability Methods. Proceedings of the ICML Workshop on Human Interpretability in Machine Learning. Stockholm, Sweden, 2018. P. 66–71.
Arrieta A. B., Díaz-Rodríguez N., Del Ser J. et al. Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. Information Fusion. 2020. Vol. 58. P. 82–115. DOI: https://doi.org/10.1016/j.inffus.2019.12.012
Banbury C., Reddi V. J., Lam M. et al. Benchmarking TinyML Systems: Challenges and Direction. Proceedings of the IEEE. 2021. Vol. 109, No. 2. P. 211–233. DOI: https://doi.org/10.1109/JPROC.2020.3031354
Caruana R., Lou Y., Gehrke J., Koch P., Sturm M., Elhadad N. Intelligible Models for HealthCare: Predicting Pneumonia Risk and Hospital 30-day Readmission. Proceedings of the 21st ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Sydney, Australia, 2015. P. 1721–1730. DOI: https://doi.org/10.1145/2783258.2788613
European Parliament, Council of the European Union. Regulation (EU) 2024/1689 Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act). Official Journal of the European Union. 2024. URL: https://eur-lex.europa.eu/eli/reg/2024/1689/oj (дата звернення: 24.04.2026).
Guidotti R., Monreale A., Ruggieri S., Turini F., Giannotti F., Pedreschi D. A Survey of Methods for Explaining Black Box Models. ACM Computing Surveys. 2019. Vol. 51, No. 5. Art. 93. P. 1–42. DOI: https://doi.org/10.1145/3236009
Hattie J., Timperley H. The Power of Feedback. Review of Educational Research. 2007. Vol. 77, No. 1. P. 81–112. DOI: https://doi.org/10.3102/003465430298487
Holstein K., Aleven V. Designing for Human-AI Complementarity in K-12 Education. AI Magazine. 2022. Vol. 43, No. 2. P. 239–248. DOI: https://doi.org/10.1002/aaai.12058
Lipton Z. C. The Mythos of Model Interpretability. Queue. 2018. Vol. 16, No. 3. P. 31–57. DOI: https://doi.org/10.1145/3236386.3241340
Lundberg S. M., Lee S.-I. A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems. Long Beach, CA, USA, 2017. Vol. 30. P. 4765–4774.
Lundberg S. M., Erion G., Chen H. et al. From Local Explanations to Global Understanding with Explainable AI for Trees. Nature Machine Intelligence. 2020. Vol. 2. P. 56–67. DOI: https://doi.org/10.1038/s42256-019-0138-9
Miller T. Explanation in Artificial Intelligence: Insights from the Social Sciences. Artificial Intelligence. 2019. Vol. 267. P. 1–38. DOI: https://doi.org/10.1016/j.artint.2018.07.007
Murdoch W. J., Singh C., Kumbier K., Abbasi-Asl R., Yu B. Definitions, methods, and applications in interpretable machine learning. Proceedings of the National Academy of Sciences. 2019. Vol. 116, No. 44. P. 22071–22080. DOI: https://doi.org/10.1073/pnas.1900654116
Redecker C. European Framework for the Digital Competence of Educators: DigCompEdu / ed. by Y. Punie. Luxembourg : Publications Office of the European Union, 2017. EUR 28775 EN. DOI: https://doi.org/10.2760/178382
Ribeiro M. T., Singh S., Guestrin C. “Why should I trust you?”: Explaining the predictions of any classifier. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco, CA, USA, 2016. P. 1135–1144. DOI: https://doi.org/10.1145/2939672.2939778
Ribeiro M. T., Singh S., Guestrin C. Anchors: High-precision model-agnostic explanations. Proceedings of the AAAI Conference on Artificial Intelligence. New Orleans, LA, USA, 2018. Vol. 32, No. 1. P. 1527–1535. DOI: https://doi.org/10.1609/aaai.v32i1.11491
Romero C., Ventura S. Educational data mining and learning analytics: An updated survey. WIREs Data Mining and Knowledge Discovery. 2020. Vol. 10, No. 3. e1355. DOI: https://doi.org/10.1002/widm.1355
Rudin C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence. 2019. Vol. 1, No. 5. P. 206 – 215. DOI: https://doi.org/10.1038/s42256-019-0048-x
Samek W., Montavon G., Vedaldi A., Hansen L. K., Müller K.-R. Explainable AI: Interpreting, Explaining and Visualizing Deep Learning. Berlin : Springer, 2019. 439 p.
Sturmfels P., Lundberg S., Lee S.-I. Visualizing the Impact of Feature Attribution Baselines. Distill. 2020. Vol. 5, No. 1. e22. DOI: https://doi.org/10.23915/distill.00022
Sundararajan M., Taly A., Yan Q. Axiomatic Attribution for Deep Networks. Proceedings of the 34th International Conference on Machine Learning. Sydney, Australia, 2017. P. 3319 – 3328.
Zimmerman B. J. Becoming a Self-Regulated Learner: An Overview. Theory Into Practice. 2002. Vol. 41, No. 2. P. 64 – 70. DOI: https://doi.org/10.1207/s15430421tip4102_2
