Integrating machine learning practicality into learning analytics: A framework for reproducible and actionable dropout prediction models

Interest in learning analytics for predicting student dropout has increased in recent years; however, a clear gap remains between research findings and their practical implementation in educational settings. This study systematically reviews 34 recent practical studies on learning analytics and dropout prediction to identify key gaps that limit the transfer of research into practice. Based on these gaps, we propose a novel six-layer framework that integrates research design, model optimization, and deployment considerations, providing a structured approach to conducting practical learning analytics research. The framework addresses critical issues, including reproducibility, generalizability, interpretability, actionability, and computational feasibility. We map the existing literature onto this framework using structured evaluation tables and find that most studies lack comprehensive attention to model practicality. Our framework contributes to the field by integrating theoretical and operational considerations at the research stage, thereby helping to bridge the gap between published research and real-world application. Furthermore, we introduce reproduction studies as a mechanism to promote innovation, particularly in improving model interpretability, generalizability, and actionability. Finally, we recommend adopting training time as a standard evaluation metric to stren then the focus on ractical feasibilit in future research.