Model strojnog učenja za detekciju kompromitiranih web-stranica u hrvatskom domenskom prostoru

master's thesis

• Title (english): Machine learning model for detection of malicious web pages in Croatian domain space
• Author: Bruno Skendrović
• Mentor: Stjepan Groš (mentor)
• Committee member: Stjepan Groš (predsjednik povjerenstva)
• Committee member: Ante Đerek (član povjerenstva)
• Committee member: Leonardo Jelenković (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2022-09-21, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Sa sve većim brojem aktivnih web stranica rastu i kriminalne aktivnosti na Internetu. Napadači putem weba pokušavaju prevariti korisnika u davanje osjetljivih podataka ili skidanja i pokretanja zlonamjernih programa. Radi sprječavanja takvih aktivnosti ovaj rad se bavi detekcijom zlonamjernih web stranica strojnim učenjem. Proučavan je skup podataka MalCrawler te su napisane skripte za dobivanje njenih značajki u svrhu stvaranja novog skupa podataka nad stranicama u hr domeni. Rezultati strojnog učenja nad MalCrawler postigli su jako dobre rezultate gdje je najbolje rezultate postigao algoritam stabla odluke C4.5 s točnošću od 0.99884 te f1 mjere od 0.97456. Servis za sigurno pretraživanje pronašao je 43 zlonamjerna URL-a hr domeni. Rezultati strojnog učenja nad hr skupom podataka postigli su jako loše rezultate te je potreban ručni pregled stranica pomoću dodatnih značajki dobivenih statičkom analizom web stranice.

Abstract (english)

With the increasing number of active websites, criminal activities on the Internet are also increasing. Web-based attackers try to trick users into providing sensitive information or downloading and running malicious programs. In order to prevent such activities, this paper deals with the detection of malicious websites using machine learning. The MalCrawler dataset was studied and scripts were written to obtain its features in order to create a new dataset over pages in the hr domain. The results of machine learning and MalCrawler achieve very good results where the best result was achieved by the C4.5 decision tree algorithm with an accuracy of 0.99884 and an f1 measure of 0.97456. The Google Safe Browsing service found 43 malicious URLs in the hr domain. The results of machine learning on hr dataset give very poor results. A manual review of the web page is required by using additional features obtained from static analysis of the web page.

• Keywords: Kibernetička sigurnost; Zlonamjerne web stranice; Strojno učenje
• Keywords (english): Cyber ​​Security; Malicious Websites; Machine Learning
• Language: croatian
• URN:NBN: urn:nbn:hr:168:887979
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 16:05:43