Exploring Multi-attribute Selection Strategies for Effective Phishing Detection with Machine Learning

Online shopping, e-commerce, and social networking have all significantly enhanced the convenience of people's job, life, and pleasure in recent years due to the Internet's rapid expansion. As a result, a growing number of individuals start using and contacting the Internet. Through phishing websites, network hackers steal confidential information from users in order to profit economically. Blacklist detection and webpage content feature detection, which are currently the most popular detection methods for phishing webpages, have the drawbacks of either being unable to identify newly developing phishing webpages or requiring manual extraction of webpage features. Feature engineering and selection is vital in phishing website detection methods, although detection accuracy is heavily dependent on prior understanding of features. Furthermore, while features recovered from several dimensions are more thorough, one disadvantage is that extracting these features takes a long time and cost. To solve these constraints, we offer a multidimensional feature extraction based phishing detection strategy based on machine learning approach. Our proposed framework works in two-phases of feature selection. In the first phase the Relief algorithm selects the most relevant feature from the URL and in the second phase the proposed Multi-Variate Attribute Selection system (MVAS) selects the features like text, code, and statistical attributes from the URL. Then finally the model is trained by combining the features selected by the Relief algorithm and MVAS system. The evaluation of the system shows that, the suggested framework has accuracy of 98.89%.