Abstract
Phishing attacks represent a significant cybersecurity threat, evolving rapidly to bypass detection mechanisms and exploit human weaknesses. This paper introduces PhishKey, a novel phishing detection framework designed to address the systemic information loss caused by data truncation in automatic methods. PhishKey employs automatic feature extraction from URL and HTML sources, combining character-level processing using a charCNN for URL classification and a Centroid-Based Key Component Extractor for word-level analysis of HTML content. Our method specifically addresses the truncation problem by ensuring complete sample processing, avoiding traditional crop operations on the input data. The predictions from both modules are integrated using a soft-voting ensemble, allowing control over the contribution of each model to the final prediction. Experimental evaluations on five state-of-the-art datasets demonstrate the effectiveness of PhishKey, achieving up to an F1-score of , and showing strong resistance to adversarial manipulations such as injection attacks, with minimal performance degradation.

This publication has 22 references indexed in Scilit: