Uma investigação do uso de características na tetecção de URLs

Abstract

Malicious URLs have become a channel for criminal activities on the Internet, such as spam and phishing. Current solutions for validation and verification of malicious URLs are considered or are believed to be accurate, with well-adjusted results. However, is it really possible or feasible to obtain 100% of accuracy in these solutions? This work describes a simple and direct investigation of features, bases and URL formats, aiming to show that the results of validation and verification URLs are highly dependent on certain aspects/factors. The idea is to extract URL features (lexical, DNS and others) for obtain the maximum information from the URLs and employ machine learning algorithms to question their influence throughout the process. In order to prove this idea, were created four hypotheses that showed that it is possible to disagree with the results of several studies from the literature.