Introducing heuristic method for combining fraud evidence based on outlier detection

Huge amount of money is lost due to fraud each year, and fraud is main obstacle for extending the electronic commerce. The dearth of labeled data is reason for conducting fraud detection studies via unsupervised methods. However unsupervised methods like outlier detection methods cannot leading up to acceptable results in fraud detection systems. In this paper heuristic method is introduced which combines the evidence gained from an outlier detection method based on their appropriate weights. At first the behavioral features of the card OWNERS ARE EXTRACTED AND PROPORTIONAL WEIGHT IS GIVEN TO EACH FEATURE Strend .then by applying fuzzy method on each behavioral trend, the outliers are detected and finally the results of the outlier detection method for each feature which is the deviation of each feature from the previous normal trends are combined according to the given weight to each feature. After applying the introduced method on real world data, we showed that by applying outlier detection on each feature and holding its result as an evidence and combining them based on their weights, rather than more accurate results the speed will improve remarkably.

Keywords:

fraud detection , fuzzy , outlier , heuristic method