Year: 2024
Type: article
Abstract: This research investigates the ability of Long Short-Term Memory (LSTM) networks to forecast financial statement fraud in publicly traded firms. Data was collected from S&P 500 companies for a decade, including 20,000 observations of company quarters. Using this data, we developed an LSTM model to identify trends in financial information across different periods.Our model takes into account approximately 50 financial indicators, including indicators related to profitability and economic health. To avoid bias in our data, we employ a method known as Synthetic Minority Over-sampling (SMOTE). We also conduct time-series cross-validation to verify the effectiveness of our tests.Results are promised. Our LSTM model outperforms traditional machine learning, achieving 95.6% accuracy, an F1 score of 0.879, and an AUC-ROC of 0.981. We see profitability, revenue performance, and revenue growth as key factors in fraud detection.Interestingly, the model's performance remained steady over different periods. It also picked up on decreasing fraud cases in recent years. This research adds to the growing body of work on AI in financial analysis and offers valuable insights for developing fraud detection methods in auditing and management.Our work shows how deep learning can uncover complex patterns of financial fraud. It lays the foundation for fraud detection and prevention in the future, potentially reshaping our approach to economic justice in the business world.
Authors: Pu, Yanli, Zhu, Yida, Xu, Haosen, Wang, Zeyu, Wei, Ming
Language: English
Subfield: Accounting
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W6968098689
FWCI:
Cites:
Cited by:
Related to: