Explainable AI Through a Democratic Lens: DhondtXAI for D'Hondt-Projected Feature Attribution
Researchers have introduced DhondtXAI, a novel framework for explainable AI (XAI) specifically designed for tabular data. This method utilizes the D'Hondt rule, a proportional representation system, to attribute feature importance. DhondtXAI offers a SHAP-independent approach that separates positive and negative evidence, allows for feature alliances, and incorporates optional thresholds. Evaluations on synthetic data and healthcare datasets demonstrate its accuracy in recovering ground truth and its high agreement with existing methods like SHAP. AI
IMPACT Introduces a new proportional attribution method for tabular XAI, potentially offering an alternative to SHAP for specific use cases.