Evaluation of attribute based phishing detection systems with machine learning methods

Phishing attacks are attacks based on deceiving users by creating fake websites that mimic legal websites, using social engineering tactics. These attacks typically involve creating counterfeit websites that imitate legal websites and often employ social engineering tactics to deceive users. Such attacks mislead users by presenting fake sites as legitimate ones to steal sensitive information. Phishing attacks, especially those aiming to steal financial data, pose a significant threat to users and the damages caused by these attacks continue to increase. Fake websites detected by the National Cyber Incident Response Center demonstrate that phishing attacks are a serious cybersecurity issue in Turkey. Effectively combating phishing attacks requires identifying their distinctive and common features before detecting such sites. This study focuses on attribute information related to URL content, character similarity, cosine similarity, Levenshtein distance similarity, and original domain features to determine if a website is a phishing site. It proposes a classifier model that achieves high accuracy rates, recognized in the literature. In this study, 32 different attribute information were used in the dataset generated within the scope of the proposed model. Three different datasets were created from the dataset prepared to detect phishing attacks within the proposed model, and the performance of various feature selection methods and classification algorithms were evaluated on these datasets. Machine learning methods such as Decision Tree, Random Forest, Logistic Regression, Support Vector Machine, and Artificial Neural Networks were employed. Tests conducted on the datasets resulted in an accuracy parameter of 96,40% for the Logistic Regression method.