Cutting-edge deep learning architectures for detecting and classifying phishing URLs

Abstract

Phishing is an increasingly prevalent cyberthreat that exploits technological and psychological vulnerabilities to result in substantial money losses and data exposures. Being alert is crucial since phishing can result in identity theft, financial loss, and security breaches. This study intends to develop and assess sophisticated deep learning models for automated phishing URL classification in order to address the swift evolution of phishing tactics and offer a reliable substitute for conventional detection techniques. The proposed model utilizes state-of-the-art architectures that combine character- and word-level embeddings with a set of hand-crafted features specifically designed to capture subtle indicators of phishing. A comparative analysis of Convolutional Neural Networks (CNN), Deep Neural Networks (DNN), and Gated Recurrent Units (GRU) reveals that CNN achieved a 96.31, DNN reached the highest accuracy of 97.81 addition of handcrafted features alongside character- and wordlevel embeddings significantly improved model performance, consistently surpassing conventional methods in all models tested. The integrating features not only enhances detection accuracy but also offers a resilient and adaptable solution to evolving phishing threats.