"Network attack detection datase"

"The datasets utilized in this experiment are sourced from CICIoMT2024 and CI-CIDS2017, which are classic experimental datasets for network attack detection [29-30].The CICIoMT2024 dataset is a IoMT dataset collected from 25 real medical devices and 15 virtual devices. The initial dataset encompasses 5 major categories and 18 sub-categories of attacks, including DDoS, DoS, scanning attacks, MQTT, and Spoofing at-tacks, among others, totaling over 8.77 million samples and comprising 44 traffic features.The initial CICIDS2017 dataset contains 7 major categories and 33 subcategories of attacks, such as DOS, brute force, web attacks, and botnets, amounting to 2.5 million samples and including 79 network flow features, such as flow duration, packet count, packet size, and flow rate.To validate the effectiveness of the algorithm proposed in this paper, we employed the Dirichlet distribution sampling method to extract a subset of samples from the complete data, serving as the training and testing sets. The sample distribution is illustrated in Tables 1-4. It is evident from the columnar data comparison that the data distribution among clients exhibits Non-IID characteristics. From a row-wise perspective, the distribution of different categories of data is highly imbalanced, with each client harboring a minority of samples. For instance, the number of class 0 data on client 1, relative to other categories, is only 71 samples, which is regarded as a minority class sample."