An improved weighted metric genetic algorithm based feature selection approach for classification of phishing URLs

Abstract

In recent years, websites have been collecting information from users for various purposes. However, users are often unaware that the collected information is being used only for its intended purpose. Phishing is a common social engineering attack that seeks to deceive users into revealing sensitive information, such as bank credentials or personal details, which can be exploited for malicious activities. Protecting this sensitive data from phishing attacks is critical in data security. Many existing approaches effectively classify phishing attacks but fail to distinguish based on the intent of the webpage. This study addresses this problem by proposing a keyword-specific web crawler focusing on extracting a novel feature similarity index based on web page content. Further, the method employs a genetic algorithm (GA) to select highly discriminating optimized features for the classification of phishing uniform resource locator (URLs). The weighted combination of different metrics is used to enhance the model accuracy. As per the performance analysis, the proposed model achieves 98.69% accuracy and 98.75% fitness value with a low false positive rate (FPR). The model interpretability is measured through Shapley additive explanations (SHAP). Additionally, the proposed method is compared with existing works, and experimental results indicate that the proposed model performance is superior over the existing solutions.