Detection of Phishing Emails

The aim of this study is to compare the effectiveness of various phishing e-mail detection methods that employ modern large language models (LLMs) alongside traditional machine-learning algorithms. Three distinct datasets are examined: one real-world corporate dataset and two publicly available datasets. The tested approaches include a custom model based on XLM-RoBERTa and Bert-Base-Multilingual, as well as SVM and Random Forest classifiers that leverage features extracted from a RoBERTa model. For comparison, a pretrained model is also evaluated. The results demonstrate that training a custom model can achieve high accuracy in phishing e-mail detection.