Detecting Accounting Fraud in Publicly Traded U.S. Firms Using a Machine Learning Approach
Yang Bao (Shanghai Jiao Tong University), Bin Ke (NUS Business School), Bin Li (Wuhan University), Y. Julia Yu (University of Virginia) and Jie Zhang (Nanyang Technological University) create a new fraud prediction model using a machine learning approach and raw financial data with a sample that covers all publicly listed U.S. firms during 1991-2008. They demonstrate the value of combining domain knowledge and machine learning method in model building. They differ from prior accounting research by using only raw accounting numbers. Using ensemble learning, a process in which multiple models strategically generated and combined, they assess the performance of fraud prediction models. Specifically, they employ one variation of ensemble learning methods called RUSBoost that makes use of random under sampling (RUS) to address the imbalance of fraudulent and nonfraudulent firms in the sample. Also, they offer a new performance evaluation metric commonly used in ranking problems that could be used by resource-constrained regulators and other monitors.
Read full paper: Bao, Y., B. Ke, J. Yu and J. Zhang. 2020. “Detecting Accounting Fraud in Publicly Traded U.S. Firms Using a Machine Learning Approach.” Journal of Accounting Research. 58 (1). https://ssrn.com/abstract=3538453