Data Mining for Fraud Detection: Toward an Improvement on Internal Control Systems?

Abstract

Fraud is a million dollar business and it's increasing every year. The numbers are shocking, all the more because over one third of all frauds are detected by 'chance' means. The second best detection method is internal control. As a result, it would be advisable to search for improvement of internal control systems. Taking into consideration the promising success stories of companies selling data mining software, along with the positive results of research in this area, we evaluate the use of data mining techniques for the purpose of fraud detection. Are we talking about real success stories, or salesmanship? For answering this, first a theoretical background is given about fraud, internal control, data mining and supervised versus unsupervised learning. Starting from this background, it is interesting to investigate the use of data mining techniques for detection of asset misappropriation, starting from unsupervised data. In this study, procurement fraud stands as an example of asset misappropriation. Data are provided by an international service-sector company. After mapping out the purchasing process, 'hot spots' are identified, resulting in some selected frauds as object of the study. As a first step towards fraud detection, outlier detection by means of clustering is practiced. The results show that this first analysis results in a good division of the sample into regular cases and cases interesting to investigate further.