MULTI-MODAL PHISHING DETECTION: INTEGRATING URL, CONTENT, AND VISUAL FEATURES FOR ENHANCED ACCURACY

Abstract

Phishing remains a critical and evolving threat in the realm of cybersecurity, exploiting user trust to extract sensitive information through deceptive websites and digital communication. To address the limitations of traditional single-modal detection systems, this project proposes a robust and intelligent Multi-Modal Phishing Detection framework that combines URL analysis, web content inspection, and visual similarity assessment. The system leverages machine learning algorithms to extract and analyze lexical features from URLs, identify structural and behavioral anomalies in web content, and detect visual mimicry of legitimate sites using screenshot-based comparison techniques. By integrating these diverse feature sets, the system achieves a comprehensive understanding of phishing patterns and enhances detection accuracy. The architecture incorporates Support Vector Machines (SVM) for URL classification, Random Forest for content analysis, and Convolutional Neural Networks (CNN) for visual feature extraction, with results fused through an ensemble-based decision model. Evaluations using a curated dataset of legitimate and phishing websites demonstrate a dummy accuracy of up to 96%, validating the effectiveness of the multi- modal approach. This research contributes significantly to the field of network security by offering a scalable, real-time phishing detection solution suitable for deployment in browsers, enterprise gateways, and cloud-based security platforms, thereby improving resilience against sophisticated cyber threats