Comparative Evaluation of Deep Learningbased Fraud Detection and Rulebased Auditing Systems Using Financial Statement Data for Ethical Compliance Assessment

— Corporate financial fraud undermines market integrity and destroys the trust of stakeholders and is a constant violation of standards of ethical compliance. In the past, regulatory bodies and external auditors have used rule-based expert systems and traditional financial ratios to identify accounting anomalies. However, these deterministic systems cannot spot non-linear and sophisticated manipulation tactics as corporate financial structures become more and more opaque and complex. In this paper, we present a comprehensive comparison between traditional rule-based auditing systems and advanced Deep Learning-based fraud detection architectures. The Autoencoder coupled Deep Neural Network is compared to industry standard rule-based heuristics with a large dataset of historical financial statements, SEC enforcement action releases and fundamental accounting ratios. Our empirical results demonstrate that the deep learning architecture achieves an F1-Score of 92.4 percent, which is a significant improvement over the 68.5 percent baseline achieved by rule-based engines. The deep learning model can effectively capture the multidimensional hidden correlations across balance sheets and income statements, and thus minimise the false positives. This provides a highly robust and scalable tool for auditors to enforce ethical financial compliance.