Web Phishing Detection Algorithm Based On Probabilistic Graphical Model

With the development of Internet and the increaseing popularity of electronic payment,Internet Fraud has gradually been the concern of public.Web Phishing(Phishing)is a way of such fraud,which uses social engineering technique through short messages,emails and IMs to induce users to visit fake website to get sensitive information like private account,token for payment,credit card information and so on.Web phishing detection is crucial for both private user and enterprises.With detecting method for phishing continually proposed and applied,the threat of web phishing has already reduced at a great extent.However,since each type of detection has limitation,phishing attackers can modify their strategies at a relatively low cost to avoid detection.Faced with the defects of current detection methods,the paper mainly focuses on the behavior pattern of phishing websites.Taking real IP flows from ISP as our subject and the results of current web phishing study as our basis,we conclude three types of feature for web phishing: Original Feature,Interaction Feature and Transfer Feature.Meanwhile,we study the probabilistic graphical model(PGM)and relevant algorithms and propose two new detection methods,they are: Detecting Model based on Markov Random Field and Belief Propagation Algorithm(BP Algorithm),and Detection Model based on Deep Belief Networks(DBN)from deep learning.The experiment suggested that our algorithm has decent accuracy and runtime efficiency.The detecting model based on BP can achieve over 90% true positive rate with less than 3% false positive rate.And the detecting model based on DBN can also achieve approximately 90% true positive rate.Both of the two detection models have the ability to detect potential phishing websites.As we have considered distributed computation while designing the algorithm,it will be easy to replicate our model in popular distributed processing frameworks.