Human-Centred Evaluation of an Interactive User Interface for Surrogate Decision Trees via Psychometrics

One of the goals of Explainable Artificial Intelligence is to enhance users’ understanding of model function and inferential capabilities by providing human-understandable explanations. An Artificial Neural Network has been trained, and interpretable decision rules have been extracted through the C4.5 algorithm. These rules were integrated into a dynamic, interactive interface that allows users to visualise and understand the inferential mechanisms behind model predictions. To rigorously assess the explainability of these rules, this research introduces a user-centred and culturally adapted evaluation, via psychometrics, of two questionnaires for XAI: the System Causability Scale and a multi-dimensional XAI scale. Findings demonstrated acceptable reliability for both questionnaires and an acceptable level of construct validity. Beyond scale translation, this research contributes to knowledge by providing a rigorously validated Italian version of existing explainability and causability questionnaires, enabling reliable cross-cultural evaluation of XAI systems and facilitating comparable empirical studies across linguistic and cultural contexts.