Student Success and Predictive Learner Analytics in Higher Education: Navigating the Ethics of Transparency

Understanding what drives learner success and how it is achieved remains a key challenge in learning analytics (LA) research[3]. Predictive LA (PLA)—a rapidly expanding research area in the field of AI in Education (AIED)—focuses on developing models to forecast future learner outcomes[6]. By analysing data such as attendance, grades, and engagement levels, AI algorithms predict learners’ likelihood of success[2]. In a time of restricted funding of higher education, the use of PLA has the potential to target students who may benefit the most from a targeted intervention, increasing the efficiency in how scare resources are allocated. Most research focuses on the accuracy of predictions, with little attention given to interpreting model decisions, explaining individual predictions to stakeholders or transparency of how decisions were made[4]. The application of AI in PLA raises ethical questions regarding privacy, fairness, robustness and transparency. A lack of transparency is considered to diminish user trust and conceal possible algorithmic biases—particularly those stemming from outdated or unrepresentative training data—which can unfairly impact specific student populations and lead to inequity of learning outcomes[7]. These risks point to a need a for human-centred learning analytics (HCLA) approach. Studies are beginning to address the issue of involving end users in the design and evaluation of LA and PLA[1]. A human-centred approach opens up the possibility for ed-tech companies, AI practitioners, policy makers, higher education leaders, staff and students and other relevant stakeholders become part of the design, deployment and implementation process. Employing a co-design approach[5] unlocks the black box of transparency, interpretability, and explainability within complex AI systems. This work in progress describes how co-design can be achieved by exploring human-centered design (HCD) methods—which can give a meaningful voice to non-technical stakeholders—and how these methods can be adopted and adapted[5]. This work focuses specifically on how a university can involve its stakeholders in the co-design, deployment, and implementation of transparent LA and PLA tools. Using a participatory design (PD) approach [9], the study applies human-centred design methods—such as interviews, journey mapping, co-design workshops, prototyping, and usability testing—to create a participatory design framework that universities can use to actively engage end users and ensure that the resulting LA and PLA tools align with their needs, values, and contexts. Within the learning analytics domain, these HCD methods play a critical role in fostering transparency, interpretability, and ethically informed decision-making.