Abstract

Emails are now extensively used across diverse domains, including business and education. However, the growing prevalence of spam poses a persistent challenge for users, leading to wasted time, resource consumption, and compromised data privacy. As spam volumes continue to rise, traditional detection techniques such as blacklists and content-based filters are proving increasingly insufficient against the evolving sophistication of fraudulent tactics. To address this issue, this study introduces a novel hybrid model that integrates a Transformer, a Multilayer Perceptron (MLP), and Bidirectional Encoder Representations from Transformers (BERT). This model is distinguished by its ability to capture the rich contextual nuances of communication, enhance classification accuracy, and detect complex patterns within lengthy texts. Its robustness and capacity to generalize to new threats were validated using a large and diverse dataset. The results indicate that the proposed model effectively balances precision and adaptability, outperforming previous approaches that often relied on limited datasets or exhibited poor generalization. Beyond serving as a classification tool, the model functions as an integrated system capable of continuous updates, making it a practical solution for improving the security of modern email systems. It achieves a high accuracy rate of 94%. In addition to underscoring the value of hybrid and advanced models in combating spam, this study provides a solid foundation for future research aimed at increasing effectiveness, improving adaptability, and minimizing the adverse impacts of spam on users and organizations.