Learning to Detect Phishing Web Pages Using Lexical and String Complexity Analysis

Abstract

Phishing is the most common and effective sort of attack employed by cybercriminals to deceive and steal sensitive information from innocent Web users. Researchers have developed major solutions to deal with this problem in recent years, but there are still a number of open challenges due to the ever-changing nature of phishing attacks. To discriminate between benign and phishing URLs, this paper proposes a static method based on lexical and string complexity analysis and distinguishing URL features. Proposed approach has been evaluated on the basis of two state of the art online learning classifiers. The confidence weighted learning classifier achieved a significant phishing URL detection accuracy of 98.35 %, error-rate of 1.65%, FPR of 0.026 and FNR of 0.005. Also, adaptive regularization of weight classifier achieved accuracy of 97.28%, error-rate of 2.72%, FPR of 0.000 and FNR of 0.052. Similar approach shows the improvement in the detection of the phishing web pages.