Explainable AI Through a Democratic Lens: DhondtXAI for D'Hondt-Projected Feature Attribution

Researchers introduce DhondtXAI, a novel explainable AI framework for tabular data that uses proportional representation principles (the D'Hondt rule) to attribute feature importance instead of relying on SHAP values. The method demonstrates high correlation with SHAP while offering complementary capabilities for handling feature interactions and alliances, validated across synthetic tests and healthcare datasets.

DhondtXAI represents an incremental advancement in explainable AI methodology rather than a paradigm shift. The research applies democratic apportionment principles to the feature attribution problem, treating feature importance allocation as a proportional representation challenge. This approach separates positive and negative evidence, enables optional feature groupings called alliances, and projects results onto local model outputs while maintaining mathematical completeness by design.

The work emerges from broader efforts to develop XAI methods beyond SHAP, which has dominated tabular data interpretation since 2017. Rather than replacing SHAP, DhondtXAI positions itself as a complementary tool that excels in specific scenarios, particularly when features interact multiplicatively. The healthcare validation on Wisconsin breast cancer and diabetes datasets demonstrates practical utility, with Spearman correlations of 0.927 and 0.935 against SHAP suggesting strong alignment on real-world problems.

For the AI development community, this offers practitioners an alternative framework when SHAP's assumptions prove limiting or when feature alliances better reflect domain logic. The diagnostic reporting of projection residuals provides transparency about information loss. However, the method's complexity—involving background-interventional removal, optional thresholding, and seat allocation—may create adoption friction compared to SHAP's widespread integration.

Investors and organizations deploying AI systems should monitor whether DhondtXAI gains traction in regulatory contexts like healthcare, where explainability requirements are stringent. The research quality and comprehensive evaluation suggest it could become a standard reference in XAI literature, though immediate market impact remains limited to academic and specialized developer communities.

• →DhondtXAI uses proportional representation principles for feature attribution, offering an alternative to SHAP-dependent methods.
• →The framework achieves exact ground-truth recovery on additive synthetic tests and dramatically reduces interaction residuals through feature alliances.
• →Healthcare validation shows 93%+ correlation with SHAP, positioning it as complementary rather than replacement technology.
• →The method maintains mathematical completeness by construction and reports diagnostic metrics for information loss.
• →Adoption depends on whether practitioners value alliance-aware and threshold-aware interpretation over SHAP's established integration ecosystem.