Supervisor: Atefeh Keshavarzi(a.keshavarzi.zafarghandi@vu.nl)
In the burgeoning realm of AI and machine learning, there is an increasing demand for comprehensible and transparent model explanations, particularly for black-box models whose predictions are often opaque. This project aims to harness the strengths of argumentation frameworks, to elucidate the predictions made by these black-box models on a single-instance basis. By modeling the decision-making process as a set of structured arguments and counterarguments, we intend to extract pertinent rules and decision pathways that can shed light on how a particular prediction was derived. These rules, derived from the intricate interplay of factors within the model, can then be presented in an understandable and intuitive manner to end-users, ensuring not only the trustworthiness of the model but also empowering users to make informed decisions based on the predictions. Such an approach promises to bridge the gap between the complex intricacies of high-dimensional models and the comprehensibility required in real-world applications.