A study about fraud detection and the implementation of SUSPECT - Supervised and UnSuPervised Erlang Classifier Tool
Alexander Lindholm · KTH Publication Database DiVA (KTH Royal Institute of Technology) · 2014
Fraud detection is a game of cat and mouse between companies and people trying to commit fraud. Most of the work within the area is not published due to several reasons. One of the reasons is that if a company publishes how their system works, the public will know how to evade detection. This paper describes the implementation of a proof-of-concept fraud detection system. The prototype named SUSPECT uses two different methods for fraud detection. The first one being a supervised classifier in form of an artificial neural network and the second one being an unsupervised classifier in the form of clustering with outlier detection. The two systems are compared with each other as well as with other systems within the field. The paper ends with conclusions and suggestions on how to to make SUSPECT perform better.