it4lia/PhishingEmailCuratedDatasets_Cleaned

Phishing Email Curated Cleaned 是一个经过清理和AI就绪的钓鱼邮件数据集，基于Champa、Rabbi和Zibran（2024）的原始“Phishing Email Curated Datasets”构建，该原始数据集聚合了11个异质邮件语料库（包括CEAS-08、Ling-Spam、Enron、Nazario钓鱼语料库、尼日利亚欺诈邮件、SpamAssassin、TREC-05/06/07以及扩展的Nazario_5和Nigerian_5分割），涵盖1995年至2022年的邮件数据。清理版本保留了数据的多源性，同时通过标准化模式、应用标签修复策略、移除重复邮件、审计常量和非有限特征，使其更易于加载、更具可重复性，并可直接用于监督分类、基于NLP的检测和跨源鲁棒性研究。数据集包含181,907个标记样本，每个样本表示为53个固定长度数值特征向量，这些特征从原始邮件文本（主题、正文、头部、URL信号）手工设计而成，而原始文本字段作为元数据单独保存以供审计和追溯。标签为二进制：0表示合法邮件，1表示钓鱼邮件。数据集适用于钓鱼邮件检测、表格ML管道基准测试、特征重要性分析以及跨源泛化研究。

Phishing Email Curated Cleaned is a cleaned and AI-ready version of the original Phishing Email Curated Datasets by Champa, Rabbi and Zibran (2024), an aggregation of 11 heterogeneous email corpora (including CEAS-08, Ling-Spam, Enron, Nazario phishing corpus, Nigerian Fraud, SpamAssassin, TREC-05/06/07 plus the extended Nazario_5 and Nigerian_5 splits) spanning 1995–2022. The cleaned release preserves the multi-source nature of the data while making it easier to load, more reproducible, and directly usable for supervised classification, NLP-based detection, and cross-source robustness studies through schema normalization, label-rescue policy, duplicate removal, and feature auditing. It contains 181,907 labeled samples, each represented as a fixed-length numerical vector of 53 features hand-engineered from the raw email text (subject, body, headers, URL signals), with original textual fields preserved separately as metadata for audit and traceability. Labels are binary: 0 for legitimate and 1 for phishing. The dataset supports phishing email detection, benchmarking of tabular ML pipelines, feature importance analysis, and cross-source generalization studies.