Enhanced Email Spam Detection Using Fine-Tuned Transformer Models and Classical Machine Learning Techniques

Email spam continues to be a pervasive challenge, causing significant disruptions to users and organizations by wasting time, reducing productivity, and posing cybersecurity risks. Traditional spam filtering techniques have achieved reasonable accuracy but often fall short when confronted with increasingly sophisticated spam tactics. Recent advancements in natural language processing, particularly transformer-based models like BERT, offer new opportunities to enhance spam detection by capturing deeper semantic and contextual information within email content. In this article presents a comprehensive comparative study of spam detection methods, combining classical machine learning classifiers with state-of-the-art fine-tuned transformer models from the Hugging Face library. We implement and evaluate standard classifiers alongside a fine-tuned BERT model for spam classification. The results demonstrate that fine-tuned transformer models outperform traditional approaches, achieving higher accuracy, precision, recall, and F1-scores while significantly reducing false positives and negatives. These improvements underline the value of applying transfer learning and deep contextual understanding to address evolving spam strategies. The study discusses the methodology for preprocessing, model training, and evaluation in detail, and includes insights into the trade-offs between model complexity and computational costs. The findings suggest that integrating fine-tuned transformer models into existing spam filtering systems can substantially improve detection robustness and reliability, paving the way for more secure and efficient email communication. Finally, we outline future directions for research, including exploration of larger transformer architectures and ensemble methods to further enhance spam detection performance.