Abstract Detecting phishing emails remains a real challenge in cybersecurity, especially as attackers are constantly finding new ways to bypass traditional defence systems. This study provides an in-depth comparison between traditional machine learning algorithms (such as Naive Bayes, Logistic Regression, SGDClassifier, XGBoost, Decision Tree, Random Forest and MLPClassifier) and more advanced deep learning models (such as LSTM, BiLSTM and GRU) in the context of phishing attack detection. We tested these models on a dataset of emails, using features extracted from both the headers and the content of the messages. The machine learning algorithms showed impressive results, with accuracies ranging from 96.01% to 98.77%. More specifically, the results were as follows: 97.92%, 98.41%, 98.77%, 97.57%, 96.01%, 98.34% and 98.77%. But on the deep learning side, performance was even better, reaching accuracies of 99.8%, 99.9% and 99.8%. This highlights the ability of these models to detect more complex attacks, using increasingly sophisticated social engineering techniques. These results underline the full potential of deep learning models for developing powerful and f lexible phishing detection systems, capable of adapting to the challenges of real-life applications.
Khayati et al. (2025) studied this question.