AI- Based Email Spam Detection

Email spam is an ongoing challenge for cybersecurity professionals.Malicious messages such as phishing attempts, malware distribution and fraudulent advertisements constantly mutate to evade detection systems.Artificial intelligence, especially machine and deep learning, has become the dominant method for spam filtering, with transformer-based models like BERT reaching detection accuracies above 98% on benchmark datasets.But existing AI-based systems are faced with a fundamental trade-off among three competing goals: detection accuracy, model explainability, and adversarial robustness.High accuracy deep models usually act as black boxes, without any explanation for their decisions, and are vulnerable to adversarial spam designed on purpose.In this paper, we provide a systematic taxonomy and comparative study of AI-based spam detection methods including classical machine learning, deep learning, transformer models, hybrid architectures and explainable AI approaches.Based on the undertaken analysis, five key areas of investigation were determined, including the absence of designlevel explainability, the lack of standardized adversarial evaluation, and multilingual spam neglect.A three-layer framework was proposed, which combines the principles of lightweight filtering, deep semantic analyses, and built-in explainability, to provide an effective theoretical basis for developing practical and dependable spam detection models.The next step in the research should be the implementation and experimentation with these ideas, as well as the conduct of adversarial benchmarking and multilingual extensions.