AI-Based Personalized Learning in Smart Education: A Systematic Review, Gap Analysis, and Framework Proposal

Abstract

Over the past decade, Artificial Intelligence technology has slowly made its way into the education sector, transforming not only teaching but also learning methods for students. However, while making such significant advances, a particular issue is overlooked – all current systems are extremely similar when it comes to personalization of lessons for students because no two learners have identical ways of thinking, pacing, or retaining information. Therefore, this paper examines ten studies published by IEEE during 2020-2025 on the topic of AI-driven personalized learning to determine the gaps within this industry. Upon analysis of the reviewed articles, common issues were discovered – poor or imbalanced datasets, recommendation engines that cannot be justified, lack of privacy measures, non-real-time adaptivity, as well as preference shown for some students over others. Based on the findings mentioned above, a proper problem statement was formulated along with a framework aimed at addressing the identified issues. The designed model is called UEPPAI-SE, and it includes a combination of multi-source student data collection, reinforcement learning engine for real-time adaptation, explanations for all recommendations, and federation of data to keep personal information from being stored on central servers.

Introduction

The text discusses the challenges and limitations of current AI-based personalized learning systems and proposes a new framework to address them.

In modern classrooms, students have highly diverse learning abilities and needs, but traditional teaching methods struggle to adapt to this diversity. While AI offers potential solutions—such as tracking student performance, identifying learning gaps, and recommending personalized content—most existing systems fail in real-world deployment. They often rely on small datasets, lack explainability, and are difficult to scale across institutions.

A review of ten research studies (2020–2025) shows that AI in education has advanced significantly through methods like collaborative filtering, knowledge tracing, deep learning, reinforcement learning, and intelligent tutoring systems. Some systems improve learning outcomes, support adaptive learning paths, and even use real-world classroom environments. However, limitations remain, including poor generalization, lack of transparency, and weak deployment readiness.

The analysis identifies five major research gaps:

1. Small and biased datasets, limiting generalization.
2. Lack of explainability, making AI recommendations hard to trust.
3. Privacy concerns, often addressed too late in system design.
4. Slow adaptation, with most systems updating after learning sessions instead of in real time.
5. Fairness issues, where systems may reinforce educational inequality.

To solve these problems, the paper proposes a framework called UEPPAI-SE, designed for explainable and privacy-preserving smart education. It introduces:

• Multi-source data integration (academic, behavioral, peer, and physiological data) for better student modeling.
• A real-time reinforcement learning adaptive engine that adjusts learning content during sessions rather than afterward.
• A transparency layer using explainable AI methods like SHAP and LIME to make recommendations understandable to teachers and students.
• A federated learning approach to protect student data while enabling decentralized model training.

Conclusion

Our goal was to examine critically the state of AI-based personalized learning without focusing on how far from reality the most hopeful papers suggest this field can go. In the end, based on ten selected papers from IEEE journal publications, what we saw was a technology with significant potential but also a series of persistent problems to solve. The progress made is indeed impressive: the successful implementation of personalized and adaptive e-learning technologies, which contribute to improving students\' academic success, increased privacy protection, advanced language teaching, which is impossible to achieve by only one instructor. However, there are still some issues to tackle: the limited scope of data used to train models, lack of explanation from models, privacy preservation measures added at a later stage, insufficiently dynamic adaptability, and preference to design technologies for convenient users. That is why the UEPPAI-SE framework that we suggest incorporates several components of personalized education technology into one integrated model, such as reinforcement learning, federated privacy preservation, knowledge graph backbone, explainability based on XAI, and fairness monitor. We consider this to be essential since solving one of the mentioned problems while others remain untouched is not enough to make any changes to the situation. There are many challenges to be addressed on the road from a framework proposal to implementation, and we have been very straightforward about those parts that have proven to be truly challenging technically, and those areas where we do not feel ethical questions have yet to be adequately addressed. It is our sincere hope that this paper will provide some clarity around what must be developed next based on a careful consideration of what has already been developed.

References

[1] [1] Y.-S. Chuang and R.-T. Hung, \"Smart learning environment for business English: The impact of AI-based tools in the classroom,\" in Proc. IEEE Int. Conf. Advanced Learning Technologies (ICALT), 2025, pp. 1–6. [2] [2] B. Pardamean, T. Suparyanto, T. W. Cenggoro, D. Sudigyo, and A. Anugrahana, \"AI-based learning style prediction in online learning for primary education,\" IEEE Access, vol. 10, 2022. [3] [3] Shashi S B et al., \"Personalised learning platform using AI-based adaptive systems,\" in Proc. IEEE Int. Conf. Inventive and Innovative Transformations in IT (ICITIIT), 2025, pp. 1–6. [4] [4] M. Murtaza, Y. Ahmed, J. A. Shamsi, F. Sherwani, and M. Usman, \"AI-based personalized E-learning systems: Issues, challenges, and solutions,\" IEEE Access, vol. 10, 2022. [5] [5] K. P. Kumar et al., \"An AI-powered smart learning platform integrating adaptive assessments, personalized learning paths, and performance analytics,\" in Proc. IEEE Int. Conf. Advances in Artificial Intelligence and Computing (ICAAIC), 2025, pp. 1–7. [6] [6] A. Ali et al., \"Personalized learning in automation: A 3D AI-based approach,\" in Proc. IEEE Frontiers in Education Conference (FIE), 2023, pp. 1–8. [7] [7] J. Zhou, \"Design of AI-based self-learning platform for college English listening,\" in Proc. IEEE Int. Conf. Machine Learning, Big Data and Business Intelligence (MLBDBI), 2020, pp. 139–142. [8] [8] Z. Feng and J. Zhang, \"Research on AI based personalized recommendation system for foreign language learning using k-nearest neighbours,\" in Proc. IEEE Int. Conf. Intelligent Systems, Communication and Networks (ICISCN), 2025, pp. 1–6. [9] [9] D. Verma et al., \"Assessing the effectiveness of AI-based learning tools in rural primary schools,\" in Proc. IEEE Int. Conf. Technology for Better Big (ICTBIG), 2025, pp. 1–5. [10] [10] Y. B. Bhushan et al., \"AI-EduAgent: A decentralized autonomous AI agent for real-time adaptive personalized learning,\" in Proc. IEEE Int. Conf. Micro-Computing Technologies and Communications (ICMCTC), 2025, pp. 1–7. [11] [11] K. T. Selvi et al., \"EduTrack: AI based personalized learning through secure analytics,\" in Proc. IEEE Int. Conf. on Intelligent Machines and Machine Intelligence (ICIMIA), 2025, pp. 1–6.

Copyright

Copyright © 2026 Om Rai, Vaibhavi Phadatare, Kritin Shetty, Tanishka Shelke, Vedant Patil, Mahi Patel, Jivesh Wassan, Vedansh Poola. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.