Continuous-Time Modeling in Educational Data Mining and Learning Analytics: A Literature Review on Methods, Ethics, and Emerging AI Trends

Abstract

This literature review consolidates key insights from Educational Data Mining (EDM) and Learning Analytics (LA), charting how AI-driven methods transform teaching, learning, and institutional decision making. Foundational studies have highlighted the potential of personalizing education, detecting at-risk students, and scaling ethical data usage. However, complexities arise from the irregular sampling of learner logs, evolving methodological frameworks, and deeply rooted concerns about equity, privacy, and interpretability. Recent discussions have introduced continuous-time modeling techniques such as Neural Ordinary Differential Equations (Neural ODEs) and Neural Controlled Differential Equations (Neural CDEs) to address irregular data streams, but empirical evidence in educational contexts remains limited. This review synthesizes foundational EDM frameworks, predictive modeling advances, equity-driven approaches, and emerging AI applications (e.g., generative AI), emphasizing their potential to reshape traditional analytics. This review also examines issues, such as fairness, stakeholder engagement, and data governance, which are critical for implementing robust and transparent analytics. By interweaving thematic areas, including socio-economic, psychosocial, and behavioral factors, this review underscores the need for interdisciplinary, ethically grounded research in continuous-time frameworks and beyond. Ultimately, these insights pave the way for a more holistic, human-centered future, where AI in education balances technical innovation with responsible equitable best practices.