About
The eXirt method was developed with the intention of providing global and local explanations of black box machine learning models. The explanations generated by this Explainable Artificial Intelligence (XAI) method are based on Item Response Theory (IRT) and the structures of the explanations consist of attribute relevance ranks and Item Characteristic Curves (ICC). The main difference between eXirt and current XAI methods present in the literature/market is the way the model is evaluated. The already known XAI methods create explanations of the model based on classical evaluations of this model (for example: accuracy, precision and recall). eXirt uses Item Response Theory, which is widely used in the evaluation of tests in educational institutions around the world, to evaluate the model according to: - Difficulty: measures whether or not the model had difficulty making the prediction; - Discrimination: measures whether the model understands the data as discriminative or not; - Guessing: measures whether the model gets certain predictions right by chance or luck.