Real Time Phishing Detection Using Lexical Analysis and Visual Similarity

Phishing, a form of social engineering, involves deceptive practices to extract sensitive information from individuals. Typically, attackers manipulate their messages to mimic genuine communication from reputable entities like financial institutions, social networks, or online marketplaces. Phishing attempts manifest across various communication channels, encompassing email, text messages, and social media platforms. The acquired sensitive data fuels identity theft and financial fraud, underscoring the importance of vigilance when encountering unsolicited communications or hyperlinks soliciting personal information or immediate action. This paper offers a comprehensive approach to tackle phishing hazards, employing diverse features including URL structure for lexical analysis and screen-shot for visual similarity using transfer learning. Notably, the XGBoostClassifier attains an outstanding accuracy rate of $$ 96.89 \%$$ . To bolster this, a hybrid approach is adopted, integrating the MobileNet model for visual similarity-based detection. The incorporation of the MobileNet model introduces a visual similarity-based detection layer, augmenting the system’s capabilities. Implemented as a Chrome plug-in, this system dynamically scrutinizes website URLs, promptly alerting users about potential phishing threats. Rigorous testing on real-world phishing websites showcases the method’s robust performance, offering a reliable and user-friendly solution to detect and prevent phishing attacks. The hybrid integration of feature based detection through XGBoostClassifier and visual similarity analysis using MobileNet fortifies the system, elevating its effectiveness in safeguarding against evolving phishing assaults.