Explainable Artificial Intelligence for Fraud Detection in Accounting: Enhancing Transparency, Accuracy, and Trust

This paper proposes a conceptual XAI-based framework for accounting fraud detection that combines accounting risk indicators, supervised fraud classification, and post-hoc explanation methods such as SHAP and LIME. The study addresses limitations of black-box machine learning in accounting by aligning model outputs with auditors' need for evidence, documentation, and professional skepticism. The proposed framework integrates data preprocessing, imbalance-aware learning, explanation validation, and human-in-the-loop review. Its main contribution is a transparent decision-support model that improves fraud-risk assessment while strengthening trust, accountability, and auditability in accounting practice.

کلیدواژه ها:

Explainable AI ، Fraud Detection ، Accounting ، Auditing ، Machine Learning