Interactive and Explainable Advising Dashboard Opens the Black Box of Student Success Prediction

This paper presents exploratory research regarding the design and evaluation of a dashboard supporting the advising of aspiring university students incorporating a black-box predictive model for student success. While black-box predictive models can provide accurate predictions, incorporating them in dashboards is challenging as the black-box nature can threaten the interpretability and negatively impact trust of end-users. Explainable Learning Analytics aims to provide insights to black-box predictions by for instance explaining how the input features impact the prediction made. Two dashboards were designed to visualize the prediction and the outcome of the explainer. The dashboards supplemented the explainer with an interactive visualisation allowing to simulate how changes in the student’s features impact the prediction. Both dashboards were evaluated in user tests with 13 participants. The results show the potential of explainable AI techniques to bring predictive models to advising practice. We found that the combination of the explainer with the simulation helped users to compare the predictive model with their mental models of student success, challenging understanding of users and influencing trust in the predictive model

ISBN:978-3-030-86435-4

Publication year:2021

Accessibility:Open