How Learning Behavior is Linked to Achievement: Bridging the Gap Between Explainable AI, Learning Analytics, and Learning Theory

Abstract

Learning behavior is a fundamental aspect of the learning process, playing a critical role in how students construct knowledge and develop skills. In educational practice, insights into students’ learning behaviors are key for effective lesson planning, early identification of those at risk, and the targeted allocation of support. As such, uncovering the relationship between learning behavior and academic achievement has been a longstanding focus in both educational psychology and learning analytics. While numerous studies across both fields have demonstrated associations between behavioral indicators and academic outcomes, important questions remain regarding how to best model this relationship in ways that are both theoretically meaningful and practically interpretable.

This dissertation addresses these challenges through two complementary approaches: (1) integrating explainable artificial intelligence techniques into academic performance prediction pipelines to produce interpretable machine learning models, and (2) embedding learning theory into the learning analytics workflow to guide feature engineering and the analysis of behavioral sequences. Although the ultimate aim is to integrate both approaches, this dissertation considers them separately in order to first demonstrate the individual validity and utility of each method. Following a substantive–methodological synergy approach, the present work advances both the analytical tools used to model the learning behavior–achievement link and the conceptual understanding of the mechanisms underlying it.

Across five empirical studies, this dissertation explores the relationship between learning behavior and achievement using behavioral trace and sensor data from digital learning environments collected in naturalistic settings and in the lab. The first study investigates the potential of explainable artificial intelligence to uncover interpretable patterns in behavioral data that predict academic performance. The second study applies explainable artificial intelligence techniques to examine individual differences in the behavior–achievement link, revealing heterogeneity in how students engage with learning tasks. The third study extends this work to sensor data in passive video-based learning settings, identifying multimodal behavioral markers that relate to learning outcomes. The fourth study introduces a systematic process for theory-informed feature engineering and compares behavioral indicators with self-reports to assess their relative predictive value. The fifth study conducts a theory-informed sequence analysis to explore the temporal dynamics of learning behaviors and their link to academic success. Key findings demonstrate that explainable artificial intelligence can uncover meaningful behavioral patterns associated with achievement and that theory-informed learning analytics provide additional explanatory power beyond purely data-driven models.

Taken together, this dissertation advances the methodological standards in academic performance prediction, proposes a blueprint for bridging educational theory and learning analytics, and contributes new insights into how specific learning behaviors relate to academic success. These contributions have implications for the design of adaptive learning systems, targeted interventions, and future interdisciplinary research in educational psychology and learning analytics.