Development of an anomaly detection model for a bank's transitory account system

Fraud detection is one of the key components of an internal control system that is designed to provide reasonable assurance regarding reliable financial statements, effective and efficient operations, compliance with applicable laws and regulations, and safeguarding of assets (AICPA 2002; Committee of Sponsoring Organizations of the Treadway Commission 1992). Practically, however, it is challenging to determine what should be investigated among myriads of a company's transactions, most of which are normal and/or free from material error. Limited human resources and cost constraints are major barriers to fraud detection in practice. This study proposes a multi-step fraud detection model that screens transactions at various levels, which leads to a manageable number of transactions that internal auditors can examine. To that end, a broad screening criterion is applied at the highest level, while more detailed criteria are used at the lowest level. This study provides an example of development of a practicable fraud detection model.