Fake Webpages Detection Using Parallel Autoencoder Neural Networks

Fake web pages attempt to steal a user's confidential information such as bank account password and email password. These fake pages are actually made similar to the pages of reputable websites such as online payment portals, Yahoo and Google, and in such a way users are drawn to these pages. This type of Internet attack is called phishing attacks. Online detection of fake pages with the help of a smart software can prevent the theft of user information and increase security in the web space. In this paper, a new approach based on autoencoder neural networks is introduced. The proposed method employs two Parallel Autoencoder (PAE) networks, one of which is trained with regular pages and the other with fake pages. At the time of detection, the type of input web page is determined based on the encoded vectors obtained from both AEs in the parallel network and a layer of conventional artificial neural network such as Softmax. In practical applications, whenever such a fake page is detected, it is promptly warned or blocked through the browser. Experimental results of the proposed method with the help of “Phishing Websites” dataset and accuracy, precision and recall criteria show that PAE networks perform better than other machine learning methods in detecting fake web pages.