SMART: Semantic, Multi-Objective, and Reinforcement-Based Adversarial Training for Email Spam Detection

This paper introduces SMART, a robust framework for email spam detection designed to adapt to and mitigate adversarial spam emails. SMART integrates several advanced techniques, including text cleaning and tokenization, word embeddings, K-means clustering for semantic enhancement, multi-objective adversarial training, and reinforcement learning for dynamic spam detection. By leveraging these components, SMART enhances the accuracy and efficiency of detecting and classifying spam emails, even in the face of evolving adversarial tactics. The framework incorporates semantic enrichment using BERT for contextual word replacements and reinforcement learning to adapt spam detection strategies dynamically. SMART provides a comprehensive solution to address the limitations of traditional spam detection methods, ensuring more secure and reliable email communication. We compared SMART experimentally with the following comparable methods: EGOAMLPs [1], MFO [2], OPTICS [3], STM [4], and MOBGOA [5]. The following is average percentage improvement of SPAM over each of these methods: 2.41% over [5]; 3.36% over [4]; 4.51% over [2]; 6.16% over [3]; 6.89% over [1].