GradCFA: A Hybrid Gradient-Based Counterfactual and Feature Attribution Explanation Algorithm for Local Interpretation of Neural Networks

Researchers introduce GradCFA, a new hybrid AI explanation framework that combines counterfactual explanations and feature attribution to improve transparency in neural network decisions. The algorithm extends beyond binary classification to multi-class scenarios and demonstrates superior performance in generating feasible, plausible, and diverse explanations compared to existing methods.

• →GradCFA combines two major explainable AI paradigms - counterfactual explanations and feature attribution - in a single framework.
• →The algorithm explicitly optimizes for feasibility, plausibility, and diversity in AI explanations, addressing key limitations in existing methods.
• →Unlike most counterfactual research focused on binary classification, GradCFA extends to multi-class scenarios for broader applications.
• →The framework shows superior performance against state-of-the-art methods including Wachter, DiCE, CARE, and SHAP in various evaluation metrics.
• →The research advances AI interpretability for critical applications in healthcare and finance where transparent decision-making is essential.