|
چکیده
|
Abstract. Fraud in the insurance industry is a prevalent issue, particularly in the form of organized schemes involving deliberate accidents and staged scenes. This paper pro- poses an algorithm designed to achieve three primary objectives. Firstly, accidents are modeled using network graph theory, with the subsequent identi cation of suspicious fraud clusters through the application of a Poisson random process. Secondly, the algo- rithm calculates the correlation between individuals involved in suspicious activity using connectivity metrics and the Monge theorem. It also examines the probability of such accidents occurring by applying local connectivity numbers in the Poisson process. This process enables the validation of each accident and individual through the assignment of a label. Lastly, while most research algorithms in fraud detection utilize data mining or arti cial intelligence, this paper overcomes the challenges posed by highly unbalanced data, including over tting and reduced accuracy.
|