Abstract
Due to the rapid expansion of Internet of Things devices and their associated network, security has become a critical concern, necessitating the development of reliable security mechanisms. Anomaly-based NIDS leveraging machine learning and deep learning have emerged as key solutions in detecting abnormal network traffic patterns. However, one challenge that affects the detection rate of machine learning or deep learning-based anomaly NIDS is the class data imbalance present in the existing dataset. Datasets are crucial for the development and evaluation of anomaly-based NIDS for IoT systems. In this study, we introduce EmuIoT-VT, a dataset generated by creating virtual replicas of IoT devices implementing a novel emulation-based method, enabling realistic network traffic generation without relying on any external network emulators. The data was collected in an isolated offline environment to capture clean, uncontaminated network traffic. The EmuIoT-VT is balanced-by-design, containing 28,000 labeled records that are evenly distributed across devices, classes, and subclasses, and supports both binary and multiclass classification tasks. It includes 82 features extracted from raw PCAP data and includes attack categories such as DoS, brute force, reconnaissance, and exploitation. This article presents the novel method and creation of the EmuIoT-VT dataset, detailing data collection, balancing strategy, and details of the dataset structure, and proposes directions for future work.