Large-Scale Automatic Classification of Phishing Pages

摘要

Phishing websites, fraudulent sites that impersonate arntrusted third party to gain access to private data, continuernto cost Internet users over a billion dollars each year. Inrnthis paper, we describe the design and performance char-rnacteristics of a scalable machine learning classifier we de-rnveloped to detect phishing websites. We use this classifierrnto maintain Google's phishing blacklist automatically. Ourrnclassifier analyzes millions of pages a day, examining thernURL and the contents of a page to determine whether orrnnot a page is phishing. Unlike previous work in this field,rnwe train the classifier on a noisy dataset consisting of mil-rnlions of samples from previously collected live classificationrndata. Despite the noise in the training data, our classifierrnlearns a robust model for identifying phishing pages whichrncorrectly classifies more than 90% of phishing pages sev-rneral weeks after training concludes.