Abstract: Phishing sites are duplicated created to trick people for their personal information, in order to simulate the real site's pages. Because they rarely detect tactical adaptability costs and identify phishing site it is a complex and dynamic problems. This work proposes a new method to detect phishing sites using Convex Programming based Transductive Support Vector Machine (CTSVM). TSVM is an independent method to detect attack, and does not change the behavior of users proposed a new way of phishing sites. Image feature extraction, and sensitive information on the page, it can completely reflect the nature of the site website. Then, phishing pages are classified by the algorithm CTSVM. Experiment result shows that the method performed well, improving the accuracy and precision are more fully. The result is to use CTSVM detect phishing sites to improve performance, such as classification TSVM should be more flexible, which makes learning CTSVM, in order to more efficiently.
Keywords: Classification, convex programming, feature selection, phishing websites, transductive support vector machine.