This paper describes an approach to fraud detection targeted at applications where this task is followed by a posterior human analysis of the signaled frauds. This is a frequent setup on fraud detection applications (e.g. credit card misuse, telecom fraud, etc.). In real world applications this human inspection is usually constrained by limited resources. In this context, standard fraud detection methods that simply tag each case as being (or not) a possible fraud are not very useful if the

more »

... er of tagged cases surpasses the available resources. A much more useful approach is to produce a ranking of fraud that can be used to optimize the available inspection resources by first addressing the cases with higher rank. In this paper we propose a method that produces such ranking. The method is based on the output of standard agglomerative hierarchical clustering algorithms, resulting in no significant additional computational costs. Our comparisons with a state of the art method provide convincing evidence of the competitiveness of our proposal.