Counterfactuals as explanations for monotonic classifiers
Year: 2023
Type: conference-paper
Abstract: Recent advances in machine learning and in particular deep learning have led to models becoming increasingly complex and less interpretable. This has led to a surge in the field of explainable AI (XAI)which aims to understand and interpret predictions made by such models. One significant direction is that of generating counterfactuals that can help in providing rich causal explanations. In this work, we present a novel counterfactual generation algorithm, with an underlying monotonic constraint respecting classifier. The generated counterfactuals are realistic and the end-user can make changes to only a few features, allowing them to make amendments easily. We demonstrate the results of our algorithm and show how this technique can generate counterfactuals closer to the query with improved coverage, while incorporating domain knowledge in the form of monotonic constraints.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7135974505