Benchmarking Model URL Features and Image Based for Phishing URL Detection

Phishing is an attack that aims to obtain someone's credentials, one of which is done by creating fake websites where the system on the website will ask users to send their personal information. A phishing detection method is developed for dealing with phishing attacks, one of which is using deep learning. The detection method utilizes deep learning and focuses on detection based on the website's Uniform Resource Locator (URL). In similar studies that have been conducted, phishing website detection methods can be carried out based on website URLs, image-based approaches through screenshots of website pages, or a combination of the two methods. This research combined text and image features as input in deep learning algorithms to detect phishing websites. Text features were obtained through URLs, while image features were obtained through screenshots of web pages. Feature extraction was done to process URLs by textual arrangement using a bag of words and finding the characteristics of URLs. The image is processed using transfer learning. This research used GRU, LSTM, and Inception V3 CNN transfer learning algorithms to create a model. The experimental trained model achieved accuracy reaching 98.2% for the combination of textual and image features, while the accuracy for textual features reached 98.8%.