AI-Driven Learning Personalization in LMS Platforms:A Systematic Review Of Mechanisms, Effectiveness, And Computational Challenges

Abstract

Background: The widespread adoption of Learning Management System (LMS) platforms in higher education has yet to overcome the fundamental limitation of uniform content delivery, which fails to accommodate individual differences in prior knowledge, learning pace, and cognitive style. Artificial Intelligence (AI) offers a transformative pathway to address this gap through data-driven personalization. Objective: This Systematic Literature Review (SLR) synthesises empirical evidence on the computational mechanisms, implementations, effectiveness outcomes, and technical-ethical challenges of AI-driven learning personalization in LMS environments within higher education, with an explicit focus on informatics and computational perspectives. Method: Adhering to PRISMA 2020 guidelines, 38 articles were selected from 312 candidates retrieved from Google Scholar, ScienceDirect, IEEE Xplore, and DOAJ (2021–2026), following three-stage screening and quality appraisal using the Mixed Methods Appraisal Tool (MMAT; minimum score 3/5). Results: Five dominant computational mechanism clusters were identified: (1) behavioral log analytics using sequence mining, clustering, and NLP; (2) academic failure prediction with Random Forest and Gradient Boosting (AUC up to 0.91, accuracy 78–89%); (3) hybrid recommender systems combining collaborative filtering, content-based filtering, and Knowledge Graph-GNN approaches (Precision@K gains of 14.3%); (4) adaptive assessment via Bayesian Knowledge Tracing combined with Item Response Theory; and (5) emerging applications of Large Language Models, Retrieval-Augmented Generation (RAG), Federated Learning, and Explainable AI (XAI/SHAP). A meta-analytic synthesis across 47 experimental studies yields a pooled effect size of d = 0.52 (medium-to-large) on academic performance. Significant challenges persist in data privacy compliance (UU PDP No. 27/2022), algorithmic fairness for 3T-region students, instructor AI literacy, and infrastructure disparity. Novelty: This review introduces a computational taxonomy of AI mechanisms in LMS, differentiating it from prior SLRs that focus predominantly on pedagogical or descriptive dimensions. Six priority research gaps are identified, including XAI adoption, culturally-fair algorithm design, and federated architectures for decentralised Indonesian institutions.