Detect social engineering attacks: Phishing Email Detection Using Deep learning and transformers for Cybersecurity
Loading...
Files
Date
Journal Title
Journal ISSN
Volume Title
Publisher
Unuversity Setif 1 Ferhat Abbas . Faculty of Sciences
Abstract
Phishing emails continue to represent one of the most dangerous cybersecurity threats due to their increasing sophistication and ability to imitate legitimate communication. Traditional detection techniques based on keyword matching and handcrafted features of-ten struggle to identify modern phishing attacks. This thesis investigates the application of deep learning and transformer models for phishing email detection, with a particular focus on BERT architectures, in addition to exploring alternative solutions like hybrid models. Several machine learning and deep learning models were evaluated, including Naive Bayes, CNN, BiLSTM, BERT, and a hybrid architecture combining BERT con-textual embeddings with TF-IDF statistical features. The models were trained and tested on a carefully preprocessed and balanced dataset containing 20,000 email samples. The proposed methodology includes preprocessing, feature representation using Word2Vec, and GloVe. Furthermore, the notion of attention mechanism was explored from imple-mentation to effect on model performance. For evaluation, metrics such as accuracy, precision, recall, and F1-score were used. As a result, the proposed BERT + TF-IDF hy-brid model produced the best overall results, achieving 99.25% accuracy and an F1-score of 99.07%, confirming that combining contextual semantic understanding with statistical keyword representation provides a more robust phishing detection mechanism. The find-ings also highlight the importance of high-quality preprocessing and feature engineering in improving model performance, and the fundamental role that attention provides in enhancing results across the board.
