Graduate School of Business
Doctor of Business Administration (DBA)
Accounting Fraud Detection, Data Analysis, Predictive Models, Machine Learning, Artificial Intelligence
Cecil, Anthony, "A Qualitative Study on Predictive Models in Accounting Fraud Detection" (2021). Doctoral Dissertations and Projects. 3243.
Companies lose an estimated 5% of revenue each year due to occupational fraud. This level of fraud can significantly disrupt the capital markets and cause companies to go bankrupt. Unless organizations, the government, and the accounting profession develop a systematic approach for accounting fraud detection, investors and employees will continue to lose money. This study explored subject matter experts’ perceptions of building and deploying artificial intelligence and predictive models to detect accounting fraud. This case study consisted of interviews with 10 participants with expertise in predictive modeling, auditing, and investigating, as well as a systematic literature review of research and technical documentation relating to artificial intelligence, predictive modeling, and accounting fraud detection. Six themes materialized from this research, including data, data mining techniques, model input and output, human agency, approach, and explainable artificial intelligence. This study attempted to expand the scope of prior research to identify the best machine learning algorithms to detect accounting fraud consistently and accurately. The results of this research revealed that labeled data is much more important to building accurate models than any machine learning algorithm or any other aspect of the model building process. However, quality labeled data is in short supply. To overcome this challenge, the research suggests using weak supervision and active learning to generate labeled data. Overall, the research shows that predictive models powered by machine learning algorithms using financial, nonfinancial, and textual data provide auditors and fraud investigators with the best chance of detecting accounting fraud.