FERB: Fraudulent camdEn geneRated phishing Benchmark

The FERB dataset was created with the CANDACE Framework [1]. To establish the contextual foundation for the framework, the town of Camden, London, was selected as a case study. Information regarding the town was gathered from publicly available sources or generated synthetically to preserve privacy and structure. The primary structural requirement was that the simulated City Council accurately mirrors an existing municipal governing body. Consequently, the content and discussions presented within the dataset do not reflect real-world events or individuals. The FERB dataset provides a total of 9,480 samples in a preprocessed format, featuring engineered attributes explicitly designed for training, validation, and testing of machine learning (ML) and deep learning (DL) models. The dataset was created at GECAD, ISEP, Polytechnic of Porto. The dataset contains the email messages and label (0 = benign; 1 = phishing). It also contains several extracted features suitable for email phishing attack detection. Some of the content of the email messages is anonymized, with tokens replacing them, enclosed between square brackets. The tokens are as follow: [EMAIL_ADDRESS], [PHONE_NUMBER], [URL], [FILE]. The emails are available in both an Apache Parquet file and a comma-separated values (CSV) file.