A Systematic Review of Self-Regulated Learning through Integration of Multimodal Data and Artificial Intelligence

While behavioral, contextual, and physiological data streams have long been used to investigate self-regulated learning (SRL), a systematic understanding of the current state how different data streams and modalities contribute to measuring regulation processes across diverse learning contexts remains limited. This systematic literature review provides a foundational step toward this understanding by addressing two objectives: (1) identifying which data streams and modalities researchers have used to capture cognitive, affective, metacognitive, and motivational (CAMM) processes underlying SRL, and (2) examining how multimodal data analytics have been applied to capture the temporal and sequential characteristics of these processes across study contexts. The studies were mapped onto the Self-Regulated Learning Processes, Multimodal Data, and Analysis grid, a two-dimensional framework with CAMM processes on one axis and multimodal data streams on the other. By evaluating four analytic approaches—unimodal, horizontal, vertical, and integrated—we identify how different combinations of data streams and processes have been employed. Our findings reveal a shift from unimodal approaches (one data stream, one process), to more integrated approaches (combining multiple data streams and processes). Although multimodal data collection is increasingly common, gaps remain, especially in measuring motivation and affective states. Futhermore, analytic methods often do not reflect alignment between data streams, with standard statistics predominating even in integrated approaches, whereas AI-based analytics may be more suited. This review positions itself as a foundational step in advancing SRL measurement, offering insights into current practices and highlighting the need for more sophisticated methods to capture SRL across diverse learning contexts.