FCX: Finding Feasible Counterfactual Explanation
Feasible Counterfactual Explanations (FCX) is a novel framework that generates realistic and low-cost counterfactuals by enforcing both hard feasibility constraints provided by domain experts and soft causal constraints inferred from data. Built on a modified Variational Autoencoder and optimized with a multi-factor loss function, FCX produces sparse, diverse, and actionable counterfactuals while preserving causal relationships, offering both individual-level explanations and global model feasibility assessments across multiple datasets _[#icdew2024]_.
References