Date of Publication :15th June 2024

Abstract:Our study methodology efficiently incorporates ma- chine learning and deep neural network technology to attack the widespread problem of phishing strategies used by fraudulent websites in the extensive digital environment. Our inquiry primarily centers on analyzing the correlation between several computational approaches, including XGBoost, SVM, Random Forest, Logistic Regression, CNN-LSTM, and CNN-BILSTM. Our technique differs from standard evaluations by concentrating on individual web pages rather than entire websites. We combine various characteristics and incorporate elements depending on the URL, domain, and content of the page. The study provides a comprehensive analysis of machine learning and deep learning methods, seeking to assess their comparative efficacy in several aspects of website functionality. This extensive evaluation encompasses a broad spectrum of prospective outcomes, guaranteeing flexibility and dependability in various circumstances. Specifically, the practical utility of our findings is enhanced by the fact that we sourced our dataset from the PhishTank website. Importantly, our work obtained a noteworthy milestone by obtaining the highest level of accuracy at 98.95% utilizing the SVM method. The exceptional precision exhibited here illustrates the efficacy of our approach in precisely detecting bogus websites. Our main aim is to help individuals and academics discover practical solutions to combat fraudulent online platforms by providing crucial comparative information. We want to contribute to the establishment of a safe online environment by exhaustively exploring the capabilities of machine learning and deep learning, particularly in the areas of fraud prevention and using various data attributes such as URL, domain, and content.

Reference :