Predicting Financial Statement Fraud with Deep Q-Network (DQN) Model: A Machine Learning Approach

Financial statement fraud (FSF) involves inflating asset, income, sales, and profit numbers while minimizing expenses, liabilities, and losses. Traditional fraud detection procedures like manual inspections and audits are costly, error-prone, and laborious. Since auditors must audit many financial statements, the system needs clever methods to detect fraud. This system proposes intelligent financial statement fraud detection via preprocessing, feature selection, and model training. Data preprocessing prepares and standardizes it. The proposed approaches select useful characteristics using the simple t-statistic technique, which comprises the t-statistic. Next, train the DQNRL model. Our proposed model outperformed RL and DQN. By outperforming these cutting-edge alternatives, the DQNRL model detected 93.42% of fake financial statements. The DQNRL-based technique detects fraud more accurately and efficiently than conventional and state-of-the-art intelligent models. Using this technology to automatically detect financial statement fraud may improve auditing processes and minimize auditor workloads.