siyam21/CIC-IoT-2023

CIC-IoT-2023物联网入侵检测数据集是来自加拿大网络安全研究所的一个数据集，经过子采样和预处理，用于机器学习评估。该数据集针对物联网环境的大规模攻击设计，包含流量级别的数值特征，用于二分类和多分类任务。数据集提供了两种配置：默认的random_3way配置（80%训练、10%测试、10%验证，采用分层随机分割，确保训练、测试和验证集完全分离）和random配置（80%训练、20%测试，用于向后兼容）。子采样策略从原始数据集中提取了1,342,314行数据，包括199,988行良性流量和1,142,326行攻击流量，覆盖33种攻击类型，以平衡二分类。特征包括39个数值流量级别特征，并列举了Top-20随机森林特征（如HTTPS、Number等）。攻击类型分为7个主要类别（如BruteForce、DDoS、DoS、Mirai、Recon、Spoofing、Web-based）和33个子类型。标签包括二分类标签（0表示良性，1表示攻击）、多分类标签（34个细粒度攻击类型）和分组标签（8个类别，包括7个攻击组和良性）。数据集没有自然时间顺序，仅提供随机分割。引用信息基于相关学术论文，许可证为CC BY 4.0。

The CIC-IoT-2023 IoT Intrusion Detection Dataset is from the Canadian Institute for Cybersecurity, subsampled and preprocessed for machine learning evaluation. It is designed for large-scale attacks in IoT environments, containing flow-level numeric features for binary and multi-class classification tasks. The dataset offers two configurations: the default random_3way configuration (80% train, 10% test, 10% validation, using a stratified random split with fully separated sets) and the random configuration (80% train, 20% test, for backward compatibility). The subsampling strategy extracts 1,342,314 rows from the original dataset, including 199,988 benign rows and 1,142,326 attack rows, covering 33 attack types to balance binary classification. Features include 39 numeric flow-level features, with a list of Top-20 Random Forest features (e.g., HTTPS, Number, Time_To_Live). Attack types are divided into 7 main classes (e.g., BruteForce, DDoS, DoS, Mirai, Recon, Spoofing, Web-based) and 33 sub-types. Labels include binary labels (0 for benign, 1 for attack), multi-class labels (34 fine-grained attack types), and grouped labels (8 classes, including 7 attack groups and benign). The dataset has no natural temporal ordering and only provides random splits. Citation is based on a relevant academic paper, and the license is CC BY 4.0.