Neural networks compared to statistical techniques

Pattern identification of stock market moves or fraudulent credit card purchases have focused on the use of statistical and neural network techniques. This project for a major credit card encompassed these two techniques in the detection of fraudulent patterns of card holder activity. The results are reported. Fraud is a crime although there are variations in its definition among the statutes of various countries where the credit card is used. Fraud is subdivided into the following categories: lost, stolen, not received, counterfeit, fraudulent application, fraudulent use of card, and other. The focus of the study is the fraudulent use of the card. Specifically, the objectives of the study were to develop a scientific approach to risk pattern analysis and to initiate development projects which significantly increase the bank´s ability to identify and control risky transaction patterns. The industry´s fraud accounts for over two billion dollars in losses each year. Although fraud losses are high in absolute dollars, they are only a small proportion of total activity. Thus, the problem of risk pattern recognition can be characterized as looking for a large number of needles in an enormously large haystack. The data used in this study include actual transactions organized by account over a six month period. Fraudulent transactions are clearly identified. This allows one to segment the accounts into good and bad meaning those that had one or more fraudulent transactions. Over fifty million transactions were used in the analysis