Machine Learning Based Phishing Website Detection System

Abstract

Attacks that are phishing related to digital financial services, mobile wallets, and online banking have risen dramatically hence putting the security and integrity of the users at a risk. Phishing sites are designed to look like legitimate online banking platforms so as to steal personal information, such as financial and user logins. Traditional phishing defense systems like rules-based and blacklist-based systems have a significant false-positive rate and are useless when negotiating new and untested phishing links. The article Phinance AI offers an analyzed phishing URL detector system based on machine learning with a particular application to banking and financial websites. The system's lightweight and real-time feature enables light detection that does not require webpage content or webpage HTML examination. The combined dataset of 483, 745 URLs obtained via publically available sources was used as the training and evaluation dataset. Lexical, structural, and entropy related characteristics were elicited because it obtained forty-one URL-based features. Several monitored machine-learning models, i.e., Random Forest, AdaBoost, XGBoost, and Gradient Boosting classifiers, were also trained and compared. The overall classification accuracy of 96.27% and heightened accuracy of 97.58% when dealing with banking related URLs give the indication that the Gradient Boosting classifier performs at a high level but with a high level of precision and a low false negative occurrence. It is a real-time phishing URL detector interface based on a web interface and API that uses Flask and Python. The results indicate that the use of machine-learning-based URL analysis is an effective and scalable process of detecting phishing with banking inclination.