Comparables XAI: Faithful Example-based AI Explanations with Counterfactual Trace Adjustments

The researchers proposed a new method called Comparables XAI, aimed at providing example-based faithful artificial intelligence explanations using counterfactual trajectory adjustment techniques. This method draws upon the comparable case technique used in real estate valuation and performs well in modeling and user research. Compared to linear regression, linearly adjusted comparable cases, or unadjusted comparable cases, Trace-adjusted Comparables achieves the highest level of explanation fidelity and accuracy, the best user accuracy, and the narrowest uncertainty boundary. Its core mechanism involves monotonically tracking the counterfactual changes from each comparable case to the target object along the artificial intelligence feature space.