Evaluating the Performance of Copula-Based Item Response Theory Models for Interpretable Assessment

Abstract

This paper describes a study evaluating the performance of copula-based Item Response Theory in real-world settings. To achieve this, we used a dataset containing information about 152 students who took a test on first-degree equations. This dataset had previously been employed to assess the performance of a Bayesian Network model in diagnosing 12 concepts related to first-degree equations. Both copula-based Item Response Theory and Bayesian Networks are explainable techniques that can be utilized for educational assessment. In this study, we compare the results of both data-driven methods against the actual state of knowledge of the students, which is a hidden variable, estimated using an expert-driven approach that involved averaging three independent assessments made by experienced primary school teachers. The results show that both methods can be used to obtain reliable estimations of students’ knowledge.