Web based Phishing Url Detection Using Deep Learning

Abstract

quotThe essence of a phishing attack revolves around harvesting sensitive data from users by fooling newlineThem into thinking that they are connecting to a legitimate website through social engineering newlineTechniques and false website interfaces. Given the evolving nature of cybercrime, the outmoded newlineDetection methods, which include blacklist-based filtering and heuristic rule-based techniques, are newlineNo longer helpful in the scanning and identification of newly created phishing attacks, referred to newlineAs zero-hour attacks. Although some work has been done to improve the situation, existing models newlineAre burdened with a lack of adaptability to novel phishing techniques and a high rate of false newlinePositive errors. In all cases, however, rude stem threats such as novel phishing techniques remain newlineHard to outmaneuver. This thesis discusses the design of an automated phishing detection system newlineThat uses deep learning systems to increase detection accuracy while striving to minimize newlineMisclassification data. newlineThe analysis opens with a detailed analysis of existing techniques implemented for phishing newlineDetection as well as their shortcomings. The deformable phishing web page is one of the primary newlineIssues that was determined. Attackers manipulate the url structure, use obfuscation techniques, newlineAnd employ domain spoofing in order to construct phishing pages that, on the surface, appear to newlineBe legitimate. To address these issues, the study proposes a bidirectional lstm with genetic newlineHeuristic algorithm and cnn(bilstm-gha-cnn) model, which is a hybrid deep learning newlineModel to classify phishing urls. It integrates bidirectional long short- term memory networks newlineWith gated highway attention and convolutional neural networks. The bilstm component newlineUtilized in our approach identifies sequential dependencies in urls, which makes it possible for newlineThe model to capture even the slightest differences and abnormalities within the data.