Odkrivanje zlorab v telekomunikacijah z uporabo analize omrežij

OBLAK, SARA (author), Šubelj, Lovro (mentor)

Zlorabe s SIM vmesnikom uporabljajo operaterji, ki s prenosom mednarodnih klicev v ciljno državo preko internetnega protokola na račun lokalnih ponudnikov pridejo do večjega dobička. V diplomski nalogi je opisan postopek zaznave zlorab s SIM vmesnikom s pomočjo različnih metod klasifikacije na podlagi podatkov o lokalnem prometu v telekomunikacijskem omrežju. Za iskanje SIM vmesniških telefonskih številk za opazovano obdobje zgradimo graf klicev med telefonskimi številkami, nato pa iz strukturnih podatkov o vozliščih sestavimo napovedni model, ki določi verjetnost, da gre pri dani telefonski številki za zlorabo. Najvišja učinkovitost je dosežena z uporabo modela naključnega gozda, dodatno pa jo povečamo še z izborom informativnih atributov. Dodatne izboljšave bi bilo mogoče doseči z analizo podatkov iz daljšega časovnega intervala ali z uporabo podatkov o trajanju klicev namesto poskusov klicev.

• Language: Slovenian
• Keywords: SIM vmesnik, detekcija zlorab, neuravnotežena klasifikacija, naključni gozd, izbor informativnih atributov
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science ; FMF - Faculty of Mathematics and Physics
• Year: 2021
• PID: 20.500.12556/RUL-130810
• COBISS.SI-ID: 78780163
• Publication date in RUL: 17.09.2021

Secondary language

• Language: English
• Title: Fraud detection in telecommunications using network analysis

Abstract

SIM box fraud is committed by telephone service providers that profit by transferring international calls to the country of the receiver through voice over internet protocol and then disguise them as local traffic. In this thesis we describe the procedure for detecting SIM box fraud by using simple classification methods on local telecommunication traffic data. To identify SIM box fraudulent numbers we build a graph of telephone numbers and calls between them for the selected time period. Further on we construct a prediction model that calculates the probability of a number's fraudulence using the node's structural data. The best performance is achieved using the random forest model that we additionally improve through feature selection. Additional performance improvements could be achieved by reviewing data collected over a longer time period or analysing data that includes the length of individual calls instead of call attempts.

• Keywords: SIM box, fraud detection, imbalanced classification, random forest, feature selection