In recent years there has been a steady spread of the so-called Fraud Detection System (FDS), fueled by an ever greater increase in malware and agents dedicated to this purpose. Various types of FDS have been developed over the years, among these the most recent refer to the branch of mathematics known as Game theory. In the work we implement a system to find and categorize fraudulent transactions in a realistic dataset provided by an Italian banking institution. The techniques on which it is based are called Follow the Belief (FB) and Follow the Regret (FR) and try to capture the type of fraud, and therefore, of the attacker by cumulatively calculating the belief and the regret over time. We then show the results and the experimental evaluations of the system on new datasets in which new synthetic transactions suitable for the purpose have been injected.
Negli ultimi anni si è assistito a una costante diffusione dei cosiddetti Fraud Detection System (FDS), alimentato da un sempre maggiore incremento dei malware e di agenti dediti a tale scopo. Vari tipologie di FDS sono state sviluppate nel corso degli anni, tra questi i più recenti si rifanno alla branca della matematica conosciuta come Game theory. Nel lavoro implementiamo un sistema per trovare e categorizzare le transazioni fraudolente in un dataset realistico fornito da un istituto bancario italiano. Le tecniche su cui si basa sono chiamate Follow the Belief (FB) e Follow the Regret (FR) e cercano di catturare il tipo di frode, e quindi, di attaccante calcolando cumulativamente nel tempo il belief e il regret. Mostriamo poi i risultati e le valutazioni sperimentali del sistema su nuovi dataset in cui sono state iniettate nuove transazioni sintetiche adeguate allo scopo.
A game theoretical approach to the fraud detection problem
CURCIO, CESARE
2021/2022
Abstract
In recent years there has been a steady spread of the so-called Fraud Detection System (FDS), fueled by an ever greater increase in malware and agents dedicated to this purpose. Various types of FDS have been developed over the years, among these the most recent refer to the branch of mathematics known as Game theory. In the work we implement a system to find and categorize fraudulent transactions in a realistic dataset provided by an Italian banking institution. The techniques on which it is based are called Follow the Belief (FB) and Follow the Regret (FR) and try to capture the type of fraud, and therefore, of the attacker by cumulatively calculating the belief and the regret over time. We then show the results and the experimental evaluations of the system on new datasets in which new synthetic transactions suitable for the purpose have been injected.File | Dimensione | Formato | |
---|---|---|---|
2022_12_Curcio.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
15.01 MB
Formato
Adobe PDF
|
15.01 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/197252