Leveraging Artificial Intelligence for Real-Time Fraud Detection in Financial Audits

Abstract

This study investigates the role of artificial intelligence (AI) in enabling real-time fraud detection within financial audits. Using a mixed-method experimental design, quantitative models—including logistic regression, random forests, gradient boosting, convolutional neural networks, and recurrent neural networks—were trained and tested on large transactional datasets. Complementary qualitative insights from audit practitioners and regulators contextualized the interpretability and ethical implications of AI adoption. Results across nine tables and twelve figures demonstrate that AI models achieved superior accuracy, precision, recall, and F1-scores compared to traditional audit approaches, with ensemble and deep learning frameworks offering the strongest classification performance. Real-time deployment simulations confirmed that fraud detection could be achieved with minimal latency and high scalability, while explainable AI techniques such as SHAP and LIME ensured model transparency. Discussion of findings emphasizes that AI enhances, rather than replaces, auditor judgment by supporting professional skepticism and reducing the audit expectation gap. The study concludes that AI-driven fraud detection contributes to stronger investor confidence, improved audit quality, and enhanced regulatory compliance, while highlighting ongoing challenges concerning ethics, data governance, and regulatory frameworks. These findings suggest that AI adoption is a critical step in advancing the future of auditing and financial accountability.