Perspectives on the challenges of generalizability, transparency and ethics in predictive learning analytics

Abstract

Educational institutions need to formulate a well-established data-driven plan to get long-term value from their learning analytics (LA) strategy. By tracking learners’ digital traces and measuring learners’ performance, institutions can discern consequential learning trends via use of predictive models to enhance their instructional services. However, questions remain on how the proposed LA system is suitable, meaningful, and justifiable. In this concept paper, we examine generalizability and transparency of the internals of predictive models, alongside the ethical challenges in using learners’ data for building predictive capabilities. Model generalizability or transferability is hindered by inadequate feature representation, small and imbalanced datasets, concept drift, and contextually un-related domains. Additional challenges relate to trustworthiness and social acceptance of these models since algorithmic-driven models are difficult to interpret by themselves. Further, ethical dilemmas are faced in engaging with learners’ data while developing and deploying LA systems at an institutional level. We propose methodologies for apprehending these challenges by establishing efforts for managing transferability and transparency, and further assessing the ethical standing on justifiable use of the LA strategy. This study showcases underlying relationships that exist between constructs pertaining to learners’ data and the predictive model. We suggest the use of appropriate evaluation techniques and setting up research ethics protocols, since without proper controls in place, the model outcome would not be portable, transferable, trustworthy, or admissible as a responsible outcome. This concept paper has theoretical and practical implications for future inquiry in the burgeoning field of learning analytics.