Lightweight AI-Driven Model for Phishing Detection with Hash-Augmented Feature Embeddings

This paper proposes a lightweight AI-driven model for phishing detection based on a hybrid embedding architecture. The model integrates standard embeddings for categorical features with a hash-augmented embedding mechanism for continuous features, enabling efficient representation with reduced computational overhead. Experimental results demonstrate a high detection accuracy of 95%, along with strong precision and recall scores. Compared to traditional models, the proposed approach achieves up to 81% fewer trainable parameters while maintaining superior convergence and classification performance, making it highly suitable for real-time and resource-constrained cybersecurity environments.