Investigating the Performance of Regression and Deep Learning Approaches to Detect Financial Statement Fraud, Focusing on Pressure/Motivation and Opportunity Dimensions

The purpose of this research is to identify the factors influencing the performance of internal audit and to evaluate the effectiveness of regression and deep learning methods in detecting fraud in financial statements, with a focus on the dimensions of pressure/motivation and opportunity. The main objectives include identifying the factors affecting pressure/motivation and opportunity in uncovering financial statement fraud using regression and deep learning models, and comparing the performance of these two approaches in simulating these factors and detecting fraud.The spatial domain of this research includes companies listed on the Tehran Stock Exchange, while the temporal domain covers the years from 1391 to 1400 (according to the Iranian calendar). Data for this study were collected using the library research method. Regression analysis was applied to investigate the factors affecting internal audit performance. Additionally, deep learning techniques, specifically feedforward neural networks, were employed for fraud detection.The results indicated that deep learning and feedforward neural network models outperformed regression methods in simulating and predicting financial statement fraud. Specifically, deep learning was more effective in capturing the relationships between the pressure/motivation and opportunity dimensions, demonstrating superior performance compared to regression models. These methods were able to uncover hidden features within the data and identify more complex factors that traditional models could not detect.This research specifically highlights the value of using deep learning techniques to identify fraud in financial statements, demonstrating that these methods can simulate and detect more complex dimensions of pressure/motivation and opportunity that were previously overlooked. It helps to fill the gaps in traditional fraud prediction models and makes a significant contribution to identifying the key factors affecting internal audit performance. Therefore, the use of deep learning can substantially enhance the accuracy and effectiveness of financial statement fraud detection methods.This research contributes to the scientific knowledge in accounting and auditing by evaluating the performance of deep learning in detecting fraud in financial statements, with a particular focus on the dimensions of pressure/motivation and opportunity. It is the first study to apply these techniques to simulate and uncover complex and hidden relationships within financial statement data. As a result, this research can assist researchers and auditing professionals in developing new and more effective tools for fraud detection.