Machine learning-based adaptive learning systems in higher education: A systematic review of techniques, outcomes, and ethical implications for knowledge translation

Abstract

Machine learning (ML) and artificial intelligence (AI) technologies have revolutionized the development of adaptive learning systems (ALS) that can personalize learning in higher education. A comprehensive and synthesized understanding of the most common ML techniques used, their reported effectiveness in improving educational outcomes, and the ethical challenges of their use in this context is needed. This systematic review synthesizes 19 peer-reviewed articles published from 2015-2025 to: (1) map and classify ML and AI techniques used in higher education adaptive learning systems; (2) assess evidence of impact on student academic performance, engagement and retention; (3) explore ethical issues such as algorithmic bias, data privacy, governance and transparency; and (4) recommend a knowledge translation approach to ethical implementation of AI-ALS within higher education settings. Following PRISMA 2020 Statement, searching in five digital databases (Google Scholar, Scopus, Web of Science, IEEE Xplore and ACM Digital Library), 19 articles were included, based on eligibility criteria, including systematic reviews, scoping reviews, empirical research and bibliometric analyses and prospective studies. The most commonly encountered ML techniques were collaborative filtering, Bayesian knowledge tracing, deep learning (LSTM and transformer-based models), reinforcement learning, NLP-based tools and learning style prediction algorithms. The studies reported significant gains in learning performance, engagement (15 - 30% improvement) and retention with AI-ALS. However, serious ethical questions were raised in most studies, such as algorithmic bias, lack of informed consent, black-box issues, and lack of institutional governance strategies, especially in under-resourced and Global South settings. ML-based ALS has great power to enhance higher education through personalized, data-driven learning. However, this potential must be realized in an ethical and equitable manner through strong governance, explainable AI, and multi-stakeholder accountability. A five-layer knowledge translation model is advocated.