Adaptive learning path generation in higher education using knowledge tracing and reinforcement learning: A multi-institutional study.
Individualized learning in higher education involves systems that modify instructional sequences to the knowledge and learning behavior of individual students. The conventional fixed curricula do not take into account dynamism of learning over time thus resulting in ineffective development and poor...
| Publicado en: | Technology & Disability p. 1 |
|---|---|
| Autores principales: | , , |
| Formato: | Journal Article |
| Publicado: |
Sage Publications Inc.
Jun2026
|
| Acceso en línea: | Ver este registro en EBSCOhost |
| fields | @attributes: recordID: 1 pdfLink: plink: https://search.ebscohost.com/login.aspx?direct=true&db=ccm&AN=194838642&site=ehost-live header: @attributes: shortDbName: ccm uiTerm: 194838642 longDbName: CINAHL Complete uiTag: AN controlInfo: bkinfo: dissinfo: jinfo: jid: 10554181 3QS jtl: Technology & Disability issn: 10554181 maglogo: N pubinfo: dt: Jun2026 pid: 344 pub: Sage Publications Inc. place: Thousand Oaks, California artinfo: ui: 194838642 10.1177/10554181261461016 194838642 ppf: 1 formats: tig: atl: Adaptive learning path generation in higher education using knowledge tracing and reinforcement learning: A multi-institutional study. aug: au: Jishuang, Long Osman, Kamisah Kutty, Faridah Mydin affil: Faculty of Education, National University of Malaysia , Bangi, Selangor, Malaysia sug: ab: Individualized learning in higher education involves systems that modify instructional sequences to the knowledge and learning behavior of individual students. The conventional fixed curricula do not take into account dynamism of learning over time thus resulting in ineffective development and poor performance. This research presents a framework of the adaptive path generation of learning based on Deep Knowledge Tracing (DKT) and Deep Reinforcement Learning (DQN) to simulate student mastery and provide the best module order. DKT model is an LSTM network that predicts the probability of mastery, based on the learning interactions in sequential form. A DQN agent makes use of these representations and designs the optimization of the learning path as a Markov Decision Process (MDP) in order to select specific modules to maximize the long-term learning gain. Multi-institutional student datasets were used to evaluate the framework based on mastery prediction, convergence, learning gain, engagement, efficiency and generalization measures. Results show the DKT model achieved 87.6% accuracy, AUC-ROC 0.7343, RMSE 0.1569, and stable convergence across 42 epochs. The DQN agent increased cumulative reward from 53.94 to 57.68, reaching 70.01 and converging by episode 742. Adaptive learning paths improved learning gain by 23.59%, reduced learning steps by 36.44%, increased average mastery from 0.68 to 0.84, and enhanced weak concept recovery to 86.5%. Engagement improved, with time spent rising from 45% to 68%, revisit rate from 12% to 32%, and dropout reduced from 18% to 5%. Cross-institution evaluation confirmed strong generalization with consistent learning gain improvements. These findings demonstrate that the framework delivers personalized, scalable learning for tutoring systems and modern education platforms, supporting diverse student populations in higher education settings. The dataset includes approximately 300 students, multiple simulated institutions, and diverse learning modules which were created from more than 9000 interaction sessions. pubtype: Academic Journal doctype: Journal Article ougenre: Article language: English refInfo: holdings: @attributes: islocal: N |
|---|