Hybrid Deep Learning and Transformer-Based Approach for Accurate Email Spam Classification

Abstract: Due to the rapid increase of unsolicited and malevolent messages, the problem of spam detection in email has become a serious issue. In this paper, we propose a comprehensive multi-model framework for spam classification integrating traditional machine learning, deep learning and transformer-based approaches. The system uses feature-based models such as TF-IDF with Logistic Regression and XgBoost, sequence based model like LSTM,CNN,BiLSTM with Attention, transformer-based DistilBERT models Experiments on a labeled email dataset demonstrate that CNN model is most performant with accuracy 98% and F1-score 0.96, BiLSTM with Attention and TF-IDF with XGBoost about 97 percent accurate. Transformer based models also provide competitive results with approx 96% accuracy. The outcome of the models underscores that hybrid and attention-based architectures are critical in improving classification performance, resisting attacks, and adjusting to changing patterns of spam.

Keywords: Email Spam Detection, Machine Learning, Deep Learning, CNN, BiLSTM with Attention, XGBoost, BERT, Text Classification, Natural Language Processing, Hybrid Models