ENHANCING PHISH-NET USING NEURALNETWORK

Phishing continues to be one of the most widespread cyber threats, exploiting deceptive websites, URLs, and emails to trick users into revealing sensitive information.Traditional detection approaches, including rulebased systems and blacklist methods, are limited in effectiveness against rapidly evolving attack strategies.This paper proposes an enhanced phishing detection framework, PhishNet++, which integrates handcrafted indicators with deep learning-based analysis to achieve higher accuracy and robustness.The system employs a character-level Convolutional Neural Network (CNN) to capture obfuscated patterns in URLs, while Transformer-based models are applied to email and webpage text for semantic understanding of phishing intent.These branches are combined with classical handcrafted features such as link counts, deceptive anchors, and suspicious keywords, forming a multi-branch architecture.The fused model delivers improved prediction performance and better resilience against zero-day attacks compared to baseline methods.Key challenges such as class imbalance, adversarial obfuscation, and real-time deployment are discussed, along with future directions including metadata and visual similarity analysis.The proposed approach demonstrates the potential of intelligent phishing detection systems to strengthen cybersecurity defenses and enhance user protection in real-world environments.