FlowCF improves sparse counterfactual explanations for mixed data
FlowCF: Sparse Counterfactual Explanations for Mixed-Type Tabular Data using Flow Matching
Machine LearningArtificial Intelligence
Summary
When trying to explain why a computer model made a certain decision, people often ask how to change the input to get a better outcome. The authors noticed that current ways to provide these 'what-if' changes usually tweak too many numbers or make big changes. They created FlowCF, a new method that finds simpler and smaller changes by treating explanations as a way to transport data points from one group to another. It works well for data with different kinds of features and keeps the changes small and focused.
What this means in practice
- •For data scientists: Generate concise and realistic counterfactual explanations for machine learning models with mixed numerical and categorical data.
- •For credit risk analysts: Provide clearer and sparser explanations on loan decisions to support regulatory compliance and customer communication.
Authors
Emmanouil Panagiotou, Eirini Ntoutsi
Abstract
In the field of Explainable AI (XAI), counterfactual (CF) explanations interpret a model's decision by suggesting the changes to the input that would lead to a more favourable outcome. To be useful in practice, such an explanation should change few features and change them as little as possible, properties known as sparsity and proximity. We observe that existing methods remain limited in this respect, especially for numerical features, whether they are model-agnostic and amortised, or gradient-based with full access to the model. In this paper, we propose FlowCF, a model-agnostic generative method that frames CF generation as sparse transport from the factual to the target class. We solve this transport with flow matching, which we extend to mixed feature types with a novel mixed flow operator, and exploit the resulting geometry to optimise for sparsity through a gating network that minimises the number of features the transport changes. Extensive experiments on six benchmark datasets demonstrate that FlowCF produces the best numerical sparsity and proximity, changing 29% of the numerical features where the best baseline changes 89%, at 70% smaller displacement, while remaining comparable on the other desiderata.