it4lia/PhiUSIIL_cleaned

PhiUSIIL Cleaned是原始PhiUSIIL钓鱼URL数据集的清理和AI就绪版本，该原始数据集是UCI机器学习仓库中的一个基准数据集，用于钓鱼URL分类，基于从URL结构和HTML页面源提取的预定义特征。它旨在检测视觉相似性钓鱼攻击（如同形异义词、Punycode、域名抢注、组合抢注），结合URL级别信号、HTML衍生信号和URL相似性指数。清理版本标准化了元数据，去除了重复项，审核了特征完整性，并导出为固定长度的50维数值特征向量，每个样本代表一个URL，同时保留原始文本列作为元数据。数据集包含235,370个标记样本，标签为0（钓鱼）和1（合法），支持二进制分类、基准测试和特征分析等用例。

PhiUSIIL Cleaned is a cleaned and AI-ready version of the original PhiUSIIL Phishing URL Dataset, a UCI Machine Learning Repository benchmark for phishing URL classification based on pre-extracted features derived from both URL structure and HTML page source. It is designed to detect visual-similarity phishing attacks (e.g., homograph, punycode, typosquatting, combosquatting) by combining URL-level signals, HTML-derived signals, and a URL Similarity Index. The cleaned version standardizes metadata, removes duplicates, audits feature integrity, and exports a fixed-length numerical vector of 50 features per sample, while preserving original textual columns as metadata. The dataset contains 235,370 labeled samples with classes 0 (phishing) and 1 (legitimate), supporting use cases such as binary classification, benchmarking, and feature analysis.