A 2-step methodology for XAI in education

The integration of AI systems in education faces significant challenges in terms of transparency and accountability. Here, we propose a two-step methodology that distinguishes between epistemic and pragmatic applications, involving human experts in the process. We conduct a case study focused on predicting low student achievement, using a large Italian dataset and employing advanced machine learning techniques. Our experimental design incorporates data-driven and theory-driven approaches within the framework of Informed Machine Learning, aiming to improve both predictive performance and explainability.