Detection of Phishing Domain Using Logistic Regression Technique and Feature Extraction Using BERT Classification Model

Abstract

Phishing attacks pose a severe danger to the security of cyberspace. Phishing is a sort of fraud in which victims and businesses are deceived into clicking on malicious URLs and disclosing private information, passwords, credit card numbers access, and other sensitive data. A method for detecting URL phishing is introduced using BERT feature extraction and a deep learning algorithm. Using BERT, the URL text was taken out of the Phishing Site Predict dataset. Phishing is a prevalent, sustainable, low-cost, and undetected network attack across the world's information networks. Using tiny-Bert to learn the semantic and long-range dependent qualities of URLs, the proposed tiny-Bert stacking based phishing website detection model extracts features from website URL strings. GBDT is the second-level learner and CatBoost, XGBoost, and LightGBM are the first-level learners in this stacking algorithm-based classifier. This detection technique may be used to identify phishing web sites without requiring the extraction of human features. Additionally, novice stackers can socially improve accuracy and generalization by making up for each other's errors throughout the categorization process. Dependent properties in URLs and get characteristics from website URL strings. The proposed classifiers, which may be used for future study, emerge as the best classifiers for the dataset utilized for phishing detection which help in global stability. In this proposed research work the detection of phishing domain is done using logistic regression and feature extraction is also done using BERT model and the results are visualized and the logistic regression shows the 96 percent accuracy. So, this will further help researchers that are already working in this domain.