Evaluating Case-Based Reasoning Knowledge Discovery in Fraud Detection

The volume of banking transaction has increased considerably in the recent years with advancement in financial transactions payment methods. Consequently, the number of fraud cases has also increased, causing billion of dollar losses each year worldwide, although from Literature, there has been substantial work in the domain of fraud detection by both the industry and academia’s. Despite the substantial work, there are few researches in applying case-based reasoning (CBR) approach in the context of detecting Financial Fraud. In this paper we aim at evaluating the performance of CBR in Identifying fraudulent patterns among financial transaction by comparing it with logistic regression (LR) and neural network (NN) which are often used in many related work. To evaluate our approach simulated data, based on a sample of real anonymous transaction provided by a bank was used and the result shows that LR outperformed NN and CBR model, with a steady increase in precision, sensitivity and specificity as the percentage ratio for the training and test data were varied. This was due to the linearity, fuzziness and presence of uncertainty in the sampling dataset. Therefore, we can reach a conclusion that part of the possible reasons why there are few research in applying CBR to the context of detecting financial fraud patterns may be due to incomplete information, fuzziness and uncertainty in the available data sets used for experimentation.