Auditors and regulators can review only a small frac-tion of companies in depth, so they need reliable ways to de-cide where to look. This study proposes and evaluates X-RBA, an explainable machine learning framework for risk-based audit analytics that ranks U.S. public companies by misstatement risk and explains each ranking in terms of audit risk areas. X-RBA combines cost-sensitive gradient-boosted trees trained on 28 raw financial-statement items with Shapley-value attributions aggre-gated into seven auditor-facing risk areas. On a public dataset of 146,045 U.S. firm-years labeled with SEC enforcement releases, under a strict out-of-sample protocol over test years 2003–2008, X-RBA achieves a mean AUC of 0.713, significantly higher than a re-implemented RUSBoost benchmark (0.675), and its hit rate among the riskiest 1% of firms is about three times the base rate. Budget-based evaluation shows that model rankings depend on the review budget. The explanations identify equity issuance and financing activity as the dominant risk drivers and show how a flagged firm’s profile can direct substantive procedures and internal-control eval-uation, supporting scalable and transparent fraud detection and regulatory compliance.