Unifying Perspectives: Plausible Counterfactual Explanations on Global, Group-wise, and Local Levels

TL;DR

The paper tackles the lack of a unified method for counterfactual explanations across local, group-wise, and global levels. It introduces a gradient-based optimization framework that jointly discovers group structure and counterfactuals, enabling end-to-end generation of Local, Group-wise, and Global Counterfactual Explanations. Plausibility is incorporated via conditional normalizing flows to ensure explanations align with the data distribution, while maintaining a balance between validity and proximity. Experimental results across six datasets and two classifiers demonstrate superior validity and plausibility, with competitive proximity, and real-world case studies (HELOC and digit transformations) illustrate actionable, group-specific recourse opportunities for diverse users.

Abstract

The growing complexity of AI systems has intensified the need for transparency through Explainable AI (XAI). Counterfactual explanations (CFs) offer actionable "what-if" scenarios on three levels: Local CFs providing instance-specific insights, Global CFs addressing broader trends, and Group-wise CFs (GWCFs) striking a balance and revealing patterns within cohesive groups. Despite the availability of methods for each granularity level, the field lacks a unified method that integrates these complementary approaches. We address this limitation by proposing a gradient-based optimization method for differentiable models that generates Local, Global, and Group-wise Counterfactual Explanations in a unified manner. We especially enhance GWCF generation by combining instance grouping and counterfactual generation into a single efficient process, replacing traditional two-step methods. Moreover, to ensure trustworthiness, we innovatively introduce the integration of plausibility criteria into the GWCF domain, making explanations both valid and realistic. Our results demonstrate the method's effectiveness in balancing validity, proximity, and plausibility while optimizing group granularity, with practical utility validated through practical use cases.

Figures (6)

• Figure 1: The figure illustrates three types of explanations generated by our approach: (a) global CFs, identifying a single direction of change applicable to the entire dataset; (b) group-wise CFs, providing vectors of change for specific groups, distinguished by colors (red, blue); and (c) local counterfactual explanations, offering instance-specific shift vectors, minimal changes required to modify individual predictions. Decision boundary (green line) and density threshold contours.
• Figure 2: Performance comparison of our counterfactual explanation framework across three granularity levels: (a) global, (b) group-wise, and (c) local, evaluated using validity, plausibility and proximity metrics. Average ranks were computed across six datasets (Blobs, Digits, Heloc, Law, Moons, Wine) and two classification models (Logistic Regression, MLP), with lower ranks indicating better performance.
• Figure 3: The figure illustrates group-wise counterfactual explanations generated using our method on the HELOC dataset with an MLP model. Each subplot highlights group-specific recommendations for financial adjustments, showing the mean change for selected financial indicators normalized over the average magnitude of changes. For each group, the number of instances is also provided.
• Figure 4: CFs for different digit pairs, showing the transformation process between different digit classes. Each row represents a distinct group. Original images are on the left, shifting vectors are in the middle column, and CFs are on the right. Red pixels in the shifting vector indicate subtracted values, while blue pixels indicate added values.
• Figure 5: Visual comparison of the efficacy of various baseline counterfactual explanation methods with our method in traversing the decision boundary of a MLP model.