CatchPhish: Model for detecting homographic attacks on phishing pages

Abstract

The growth in the numbers of phishing attacks, along with the volume of successful frauds, demonstrates vul-nerabilities of the protection tools and exposes the advance in the refinement of the attacks. In more than 70% of cases, the improvements rely on the presence of homographic terms as a mechanism to embed reliability in malicious pages. In this scenario, the present study proposes an intelligent approach denominated CatchPhish, which, through the attack target brand identification, can infer the veracity of the page evaluated. CatchPhish uses a Siamese neural network capable of identifying the presence of typosquatting mentions in phishing pages. In the experiments, the proposed approach achieved 99.30% of assertiveness. In addition, the proposed approach stands out for its ability to produce terms for training, so, instead of providing the tool with a high amount of distorted terms, it provides the mark preceded by the correct spelling, which circumvents a strong obstacle in the construction of protection mechanisms.