Comparative Effectiveness of Adaptive Learning Platforms in STEM Education: A Systematic Review

US Journal of New Insights in Tech & Education

Kanita Haider, Md Rasel Ul Alam, Oishe Al Mariz

Chittagong University of Engineering and Technology, Chattogram 4349, BD; University of the Cumberlands, Kentucky, USA

US Journal of New Insights in Tech & EducationVol. 4, Issue 1July 25, 2024

View Full PDF

Abstract

Adaptive learning platforms (ALPs) use algorithmic models of learner performance to individualize the sequence, pace, and difficulty of instructional content, and have been widely adopted across science, technology, engineering, and mathematics (STEM) disciplines. This systematic review synthesizes experimental and quasi-experimental evidence on the comparative effectiveness of ALPs relative to non-adaptive instruction on STEM learning outcomes. Following PRISMA 2020 guidelines, 1,842 records were identified across five databases; after deduplication, screening, and full-text assessment, 58 studies (112 extracted effect sizes; approximately 46,000 learners) met inclusion criteria. A random-effects meta-analysis produced a pooled effect of g = 0.63, 95% CI [0.51, 0.75], favoring adaptive over non-adaptive instruction, with a larger effect in undergraduate STEM contexts (g = 0.85) than in primary or secondary contexts. Subgroup analyses indicated the strongest gains in mathematics and computing disciplines, moderate gains in science, and comparatively smaller gains in engineering. A supplementary machine-learning-based moderator analysis (MetaForest, a random-forest adaptation for meta-analysis) ranked STEM discipline, educational level, and intervention duration as the most influential moderators of effect size, ahead of sample size and study design. A domain-level risk-of-bias assessment and leave-one-out sensitivity analysis, reported here in an expanded results section, indicated that the pooled estimate was not driven by any single study or by studies at elevated risk of bias. Funnel-plot and trim-and-fill diagnostics suggested modest asymmetry but did not substantially alter the pooled estimate. The findings support the instructional value of adaptive, data-driven personalization in STEM contexts while underscoring the need for longer intervention windows, algorithmic transparency, and attention to equity of access across educational levels.

Keywords

Adaptive Learning STEM Education Systematic Review Meta-Analysis Machine Learning Educational Technology Risk of Bias

Article Information

Published
July 25, 2024
Journal
US Journal of New Insights in Tech & Education
Volume / Issue
4 / 1
Article No.
USJNITE- 2401
Year
2024

Browse

All published articles · Journal archive