Explainable Fuzzy Models for Learning Analytics

Abstract

Learning Analytics has been widely used to enhance the educational field employing Artificial Intelligence. However, explanations of the data processing have become mandatory. In order to do this, we suggest using Neuro-Fuzzy Systems (NFSs), which can provide both precise forecasts and descriptions of the processes that produced the outcomes. The balancing between model explainability and accuracy has been studied by reducing the number of relevant features. Click-stream data, describing the interactions of the students with a Virtual Learning Environment has been analyzed. Results on the OULAD datset have shown that the NFS model provides effective prediction of the outcomes of students and can well explain the reasoning behind the prediction using fuzzy rules.