CyberFence: Intelligent Defense Against Phishing Links

Abstract

The project addresses the growing threat of phishing and malicious websites, which cause financial loss, identity theft, and distrust in online services. It proposes a real-time URL classification system that integrates lexical, structural, behavioral, and reputation-based features. By leveraging traditional ML baselines (Random Forest, Naïve Bayes) along with a Residual Multi-Layer Perceptron (ResMLP) model, the system achieves high accuracy (~95%) and low inference latency (~50ms). An interactive dashboard enhances interpretability, ensuring trust in predictions and suitability for deployment in high-throughput environments.