A New Hybrid Fraud Detection Approach in Credit Cards and Financial Statements

This paper presents a new method for fraud detection in credit cards and financial statements. This method uses a combination of three regular methods i.e. Support Vector Machine (SVM), Neural Network and Decision Tree (DT). After training, each mentioned structure is given with samples to make appropriate labels. These labels are combined separately with three other classifiers to achieve final label which indicates the presence or absence of fraud. In addition, T-statistic feature selection is used for preprocessing as well as an approach based on Euclidean and middle distance for data balancing. To evaluate proposed method, we have benefited from two data sets of UCI database and data sets of 2nd International Artificial Intelligence and Robotic Competition which was held in 2013 at Amir Kabir University in Iran. Efficiency of the proposed method on balanced and unbalanced discussed and Mann-Whitney testshows a significant difference compared with the other methods and this approves the performance of the proposed method. Improving in performance, increasing in response speed and decreasing false alarms are the major achievements of the proposed method.

Keywords:

Fraud detection , Balancing , Support vector machine , Neural network , Decision tree