PACE: A Neuro-Symbolic Framework for Plausible and Actionable Counterfactual Explanations
06:00 · July 3, 2026 · arXiv cs.AI RSS
Counterfactual explanations explain machine learning predictions by identifying minimal input changes that would alter a model's decision. Although many existing methods successfully generate prediction-changing alternatives, they often produce unrealistic or infeasible recommendations due to a lack of explicit mechanisms for incorporating domain knowledge and intervention constraints. Neuro-symbolic AI offers a promising direction by combining data-driven predictive models with symbolic reasoning capable of representing human-understandable rules and feasible actions. This paper presents PACE, a modular neuro-symbolic framework 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, the framework produces explanations consistent with domain knowledge while remaining interpretable and actionable. The approach is model-agnostic and adaptable to domains requiring realistic decision support. A case study is conducted on the Adult Income dataset, combining a multilayer perceptron classifier with Answer Set Programming (ASP) rules encoding feasible modifications to education, occupation, and working hours while preserving immutable attributes. Results highlight the trade-off between counterfactual validity and plausibility and show that symbolic constraints yield explanations that better satisfy domain-specific feasibility requirements, illustrating the potential of neuro-symbolic methods for transparent, feasibility-aware counterfactual explanation in explainable AI.
Summary
Counterfactual explanations aim to clarify machine learning decisions by identifying the smallest changes to an input that would flip a model’s output. In practice many current generators optimize only for validity and proximity, which frequently yields recommendations that violate real-world constraints or cannot be acted upon. PACE addresses this limitation through a modular neuro-symbolic architecture that pairs a neural classifier with a symbolic reasoning layer based on Answer Set Programming. The neural component handles prediction while the ASP layer encodes admissible interventions and domain rules, restricting the search to feasible states and preserving immutable attributes.
The framework is model-agnostic, allowing any trained classifier to be coupled with a domain-specific knowledge base that defines which variables may change and under what conditions. In the reported case study a multilayer perceptron is applied to the Adult Income dataset; ASP rules govern modifications to education level, occupation and working hours while treating attributes such as age, sex and race as fixed. Candidate counterfactuals are therefore generated only within the space of realistic actions, producing explanations that remain both prediction-changing and consistent with domain knowledge.
Empirical comparison against optimization-based and generative baselines (including DiCE, Wachter, VCNet and C-CHVAE) illustrates the inherent trade-off between counterfactual validity and plausibility. When symbolic constraints are enforced, the proportion of feasible outputs rises markedly, although the distance to the original instance can increase. The results indicate that explicit feasibility modeling improves actionability without sacrificing interpretability, offering a practical route toward trustworthy decision support in regulated domains.
Why it matters
This research is highly relevant to the Dutch AI market's strong emphasis on ethical, transparent, and GDPR-compliant AI. The neuro-symbolic approach to explainable AI (XAI) provides researchers and advanced practitioners with actionable methodologies to build interpretable systems that respect real-world constraints.
