Do student advisors prefer explanations using local linear approximations (LIME) or rules (LORE) in the prediction of student success?

Student advisors want to provide aspiring students with appropriate recommendations or remedial actions in their transition from secondary to higher education. A prediction of student success might provide support in this process. However, as student success prediction models are often black box models, there is a need for explainable AI that helps student advisors to understand why a student has a higher chance of failure or success. Understanding why a black box model makes a certain prediction is also important in assessing trust, which is fundamental when actions will be taken based on the model’s predictions. This work makes a comparison between different explainers supporting a black box prediction of student success. The comparison focuses on two model-agnostic explainers: Local Interpretable Model-agnostic Explanations (LIME) and LOcal Rule-based Explanations (LORE). The poster will present an evaluation with end users (student advisors). The results indicate that advisors prefer working with a LIME explanation over working with a LORE explanation because of the extra nuanced information in the LIME explanation.

Jaar van publicatie:2021

Toegankelijkheid:Open