Preprocessed MQTTset for IoT Network Intrusion Detection (MLOps-Ready Dataset)

This dataset contains preprocessed network traffic derived from the publicly available MQTTset dataset for IoT intrusion detection. Specifically, version 5 (released on 2021-07-14) was used. The data has been transformed to support reproducible machine learning and MLOps workflows.

Key preprocessing steps include:

Synthetic temporal interleaving of attack packets within normal traffic to simulate realistic deployment conditions

Generation of multiple datasets for different stages of the ML lifecycle:

reference.csv: baseline normal traffic for creating pre-training tests

training.csv: mixed traffic with randomly injected attacks for model training

post-deployment.csv: simulated production traffic with randomly injected attacks for evaluation

The dataset is intended for:

IoT intrusion detection model development

Data validation and monitoring experiments

Reproducible MLOps pipelines

Notes:

Attack timestamps are synthetically generated and do not reflect original capture timing

The dataset preserves the temporal structure of legitimate traffic

The previous version of this record used an incorrectly constructed training.csv.

本数据集包含经预处理后的网络流量数据，源自面向物联网（IoT, Internet of Things）入侵检测任务的公开可用MQTTset数据集，本次使用的是2021年7月14日发布的第5版。该数据已完成适配处理，以支持可复现的机器学习（ML）与机器学习运维（MLOps）工作流。

关键预处理步骤包括：

在正常网络流量中对攻击数据包进行时序交错合成，以模拟真实的部署运行条件

针对机器学习生命周期的不同阶段生成多份数据集：

reference.csv：用于开展预训练测试的基准正常流量数据集
training.csv：包含随机注入攻击流量的混合数据集，用于模型训练
post-deployment.csv：带有随机注入攻击流量的模拟生产环境流量数据集，用于模型评估

本数据集适用于：

物联网入侵检测模型开发
数据验证与监控实验
可复现的机器学习运维工作流管道

说明事项：

攻击时间戳为合成生成，不反映原始捕获的时间节点
本数据集保留了合法流量的时序结构
本记录的过往版本使用了构建有误的training.csv文件。