Проверка подлинности сайта организации с использованием методов машинного обучения

Abstract

The analysis of news resources confirms the presence of financial damage caused to users by threat implications, the sources of which are the organisation’s fake websites. As the problem solution of the organisation’s fake websites, the article discusses methods for authenticating web resources. For this, a number of features characteristic of the organisation’s original sites are identified. The technique of forming a sample of the organization’s genuine sites using information and analytical systems for business is presented. On a training sample containing sites from list-org.com and the PhishTank service, models based on applying the Bayes theorem, a decision tree, and a separating hyperplane are carried out. A histogram is provided showing the proportion of sites for each of the samples that have the corresponding authentication criterion. When implementing programmes based on using these methods, the Java programming language and the Weka library are used. The effectiveness of these models is evaluated on a test sample containing other sites from the same resources. The separating hyperplane allows having a higher overall classification accuracy. At the same time, when using the naive Bayes classifier, the least number of errors are made when the fake sites are classified as genuine.