Domain Name-Based Phishing Detection Using Deep Learning Techniques.

Phishing continues to be one of the most prevalent cyber-attacks, deceiving users into revealing sensitive information through malicious domains and fraudulent websites. Traditional phishing detection methods rely heavily on full-URL inspection, blacklist matching, or web-content analysis—approaches that are computationally expensive, slow, and ineffective against zero-day attacks. This paper proposes a lightweight and domain-centric phishing detection system using Artificial Neural Networks (ANN). The system analyzes only domain-level features, making it faster, more scalable, and suitable for real-time applications such as browser extensions. A dataset of over 52,000 domains (balanced between legitimate and phishing) was used, from which 20 WHOIS-based features and 44 content-derived features were extracted. A custom ANN model trained on these 64 combined features achieved 88% classification accuracy. To ensure practical deployment, the model was integrated into both a web application and a browser extension, enabling real-time domain safety verification with minimal latency. Additionally, a user-feedback mechanism supports continuous retraining, helping the model adapt to evolving phishing strategies. Beyond its technical contributions, the proposed system demonstrates strong potential for real-world cybersecurity integration. Because the model relies exclusively on domain-level metadata, it operates independently of webpage content, JavaScript execution, or visual features making it suitable for low-resource devices, mobile platforms, and privacy-sensitive environments. The lightweight architecture ensures rapid inference times, while the modular design allows for seamless integration into browsers, enterprise security systems, and cloud-based threat filters. Overall, this work provides a scalable, efficient, and proactive defense mechanism capable of mitigating modern phishing threats before users interact with malicious webpages.