Defining Adaptive Whitelists by Using Clustering Techniques, a Security Application to Prevent Toll Fraud in VoIP Networks

摘要

One strategy to prevent telephone fraud used by telephone operators is the so called White list which is a set of international destinations where the user is allowed to call, it is by default static and is changed manually by request of the user. This paper describes the creation of adaptive White list that will change based on the VoIP traffic behavior of corporate customers, starting with a list of countries based on the frequency of calls to every destination per customer, so an easy way to create a small set of Whitelists rather than a White list per customer, but equally effective to block fraud calls, is proposed. As well the Whitelists are adaptive so they are changing according to user group's behavior over the time. The Weka's SimpleKMeans method is used to cluster international destinations where the customers use to call on a six month period. We describe briefly the K-means method and a way to measure its effectiveness to generate clusters of destinations per customer, the validation method includes the EM algorithm which is briefly described as well. The generated clusters are tough to feed a machine learning system used to detect toll frauds to be developed as a next phase.