Abstract
Counterfactual explanations elucidate machine learning predictions by pinpointing minimal alterations to input features that would change a model's decision. While many existing methods successfully generate such prediction-changing alternatives, they often produce unrealistic or infeasible recommendations due to the lack of explicit mechanisms for incorporating domain knowledge and intervention constraints. Neuro-symbolic AI presents a promising direction by merging data-driven predictive models with symbolic reasoning that can represent human-understandable rules and feasible actions.
This paper introduces PACE, a modular neuro-symbolic framework designed for generating feasibility-aware counterfactual explanations. The framework separates prediction and reasoning into two components: a neural predictive model for classification and a symbolic reasoning layer that enforces domain-specific constraints during counterfactual generation. By explicitly modeling feasible interventions, PACE yields explanations consistent with domain knowledge while remaining both interpretable and actionable. The approach is model-agnostic and adaptable to domains requiring realistic decision support.
A case study on the Adult Income dataset demonstrates the framework's effectiveness, combining a multilayer perceptron classifier with Answer Set Programming (ASP) rules. These rules encode feasible modifications to education, occupation, and working hours while preserving immutable attributes. Results highlight the trade-off between counterfactual validity and plausibility, revealing that symbolic constraints produce explanations better aligned with domain-specific feasibility requirements. This work underscores the potential of neuro-symbolic methods for transparent, feasibility-aware counterfactual explanation in explainable AI (XAI).
arXiv: 2607.01306 (cs.AI)
DOI: 10.48550/arXiv.2607.01306
via ArXiv AI