Machine Learning Applications in Forensic Accounting: Detecting Financial Statement Fraud Through Neural Network Analysis

Abstract

This research investigates the application of machine learning techniques, speci}cally neural networks, in detecting }nancial statement fraud
within forensic accounting. The study develops a comprehensive fraud
detection model using }nancial ratios, transactional patterns, and corporate governance indicators from 1,200 publicly traded companies spanning
2010-2023. Our methodology employs a multi-layer perceptron neural network architecture trained on validated fraud cases identi}ed by regulatory
authorities. The model achieves 94.2% accuracy in fraud classi}cation, signi}cantly outperforming traditional statistical methods. Key predictive
variables include abnormal accruals, related-party transaction frequency,
and board independence metrics. The research demonstrates that machine learning approaches can substantially enhance fraud detection capabilities in accounting practice, providing auditors and regulators with
more ezective tools for }nancial crime prevention.