Detection of credit card fraud by using support vector machines and neural networks

Conventionally, historical actual transaction data are. used to set up a model for detecting credit card fraud. Instead of using traditional approaches, a new personalized approach has recently been presented to prevent fraud. The personalized approach proposes to prevent credit card fraud before initial use of a new card, even users without any real transaction data. Though this approach is promising, there are still some problems waiting to be improved. A main issue of the personalized approach is how to predict accurately with only few training data, since it collects quasi-real transaction data by using an online questionnaire system and users are generally not willing to spend too much time to answer questionnaires. This study employs support vector machines (SVM) and artificial neural networks (ANN) to investigate the problem of fraud detection of credit cards. The type of ANN models we use in this study is the back propagation networks (BPN). The performance of neural networks is compared with that from SVM. Experimental results from this study show that both BPN and SVM can offer good solutions. When the data number is small, BPN can have better prediction performance than SVM. Besides, the average prediction. accuracy reaches a maximum when the training data ratio arrives at 0.8.