Modelling audit risk with AI and explainability: Cross-country evidence from emerging and mature markets

Abstract

This study examines how artificial intelligence (AI) compares with traditional econometric models in predicting audit risk across two institutional contexts: the United Arab Emirates (UAE) and the United Kingdom (UK). Using firm-level data from 2017–2024, audit risk is modelled using financial, governance, audit, and market factors. Logistic and probit regressions serve as econometric benchmarks, while Random Forest, XGBoost, and deep neural networks represent AI methods. Explainable AI (XAI) tools—such as SHAP and LIME—enhance interpretability and regulatory transparency. Findings show that AI models consistently outperform econometric ones in accuracy, recall, and AUC across both countries. Yet, key audit risk drivers vary: governance factors like board independence and ownership concentration dominate in the UAE, while financial indicators and Big 4 affiliation are more influential in the UK. Explainability tools clarify predictions, boosting trust and regulatory alignment. Cross-country transferability tests reveal lower accuracy outside the original setting, emphasizing institutional specificity. Overall, the study demonstrates that effective audit risk prediction requires locally adapted, transparent AI models that combine predictive strength with interpretability to enhance auditor and regulator confidence.