From Financial Auditing to Algorithmic Auditing : Toward an Ethical and Fair Assessment of Credit Risk in the AI Era, Riham Benamor, Dr Azhaar Lajmi

Riham Benamor
PHD Student
University of Tunis, GEF2A-Lab
High Institute of Management of Tunis
Tunisia 
Dr Azhaar Lajmi
Associate Professor
University of Tunis, GEF2A-Lab
High Institute of Management of Tunis
Tunisia

Abstract: This research proposes a transposition of the IFRS 9 logic to the algorithmic audit of AI models used for credit scoring. Several machine learning algorithms, including logistic regression, decision trees, Random Forest, and XGBoost, were tested and compared on the Credit Risk Dataset available on Kaggle. The evaluation is based on predictive performance, but also on bias detection using fairness metrics such as Disparate Impact and Demographic Parity. In summary, XGBoost seems to deliver the best accuracy. Nevertheless, we still observe a strong weighting of sensitive variables (age, property ownership status) in credit decisions. Our study stands out due to the intersection between the frame of IFRS 9 and the ethical audit of artificial intelligence algorithms, two rarely connected fields. Overall, it paves the way for technological regulation aligned with international transparency standards, promoting responsible AI and ethical credit risk assessment. The proposed framework supports responsible AI adoption in financial institutions and contributes to the development of audit practices aligned with IFRS 9 and Basel III transparency requirements.
Keywords: Audit, IFRS 9, Explainable AI, Fairness, Credit Risk, FinTech, Ethics.