Phishing link detection system

Phishing attacks continue to pose a serious cybersecurity threat by simultaneously exploiting technical weaknesses and human psychological behavior to obtain sensitive information. Existing detection approaches are largely limited to single-dimensional analysis, such as examining suspicious URLs or email content in isolation, which reduces their effectiveness against modern, adaptive phishing techniques. To overcome these limitations, this project presents a multimodal phishing detection system that performs a comprehensive analysis across multiple dimensions. The proposed approach integrates URL-based feature extraction, advanced natural language processing of textual content using models like linear regression, random forest and XGBoost algorithm as ensemble model, and psycholinguistic analysis to capture social engineering indicators such as urgency, fear, and perceived trust. Models are combined to improve classification accuracy, robustness, and generalization. By leveraging complementary strengths from multiple models and feature modalities, the system demonstrates superior adaptability compared to traditional single-feature detection mechanisms. The final outcome is a practical, real-time protection solution implemented as a browser extension that automatically evaluates websites upon loading and an additional module capable of analyzing email text to detect phishing attempts, thereby offering an effective and user-oriented defense for everyday online interactions.