This research paper presents PrivRecourse, a solution to the privacy issues that arise when generating realistic recourse paths based on instance-based counterfactual explanations. The system uses differentially private (DP) clustering to represent non-overlapping subsets of the private dataset. These DP cluster centers are then used to generate recourse paths, resulting in realistic and actionable paths. The approach was evaluated on finance datasets and compared with other methods, demonstrating that PrivRecourse can provide private and realistic paths.
Publication date: 2023-11-27
Project Page: ?
Paper: https://arxiv.org/pdf/2311.14137
