Phishing Website Detection Via Shuffle Net V2 And Exemplar Pyramid Deep Feature Extraction Network (EPDFE)

Customers are moving from traditional retail to online shopping due to the Internet's tremendous growth. Instead of robbing banks or shops anymore, modern criminals use a variety of unique strategies to locate their victims online. Due to the invisibility of the Internet, hackers have developed new strategies, such as phishing, to trick victims into entering sensitive information, such as usernames, passwords, and account IDs, on fake websites. Detecting if a web page is accurate or phishing is a very challenging task because of the semantics-based assault structure of phishing, which primarily preys on the vulnerabilities of computer users. Software developers frequently create new anti-phishing innovations, however, even when these tools employ blacklists, heuristics, visual and machine learning-based techniques, they can't stop all phishing attacks. Several ML based phishing detection methods have been proposed due to the development and use of ML technology. As a result of specific solutions depending on features extracted by rules and others needing to rely on other services, the prediction service may encounter instability and time-consuming issues. This paper provides the deep learning-based shuffle Net V2 approach for phishing website detection, which is optimized by an enhanced crow search optimization algorithm (ECSO). The feature extractor receives a list of website URLs and uses an exemplar pyramid deep feature extraction network (EPDFE) to extract the crucial features. The collected features are given to the Enhanced rat swarm optimization (ERSO), which chooses the features. The proposed method performs well in actual phishing discovery, indicating its great applicability in performances and enhancing detection outcomes while significantly reducing execution time.