MU-IoT: A New IoT Intrusion Dataset for Network and Application Layer Attacks Analysis

As the growth of IoT networks increases exponentially, the number of cyber attacks is also increasing on IoT networks day-by-day. This results in the vital requirement of cyber security mechanisms to secure IoT networks from cyberattacks. To build such a security mechanism, researchers and cybersecurity practitioners need relevant IoT datasets. However, until now, only a few publicly available IoT network intrusion datasets exist in the literature. Moreover, these datasets lack coverage of IoT network traffic containing IoT application layer protocols like AMQP, XMPP, STOMP, etc. and various IoT-based attacks. So, in this work, we generate a new realistic and comprehensive IoT network intrusion dataset called MU-IoT for IoT cybersecurity. The network traffic contained in the MU-IoT dataset is collected from the Manipur University (MU) IoT Smart Lab, which is our own IoT network testbed. The testbed includes more than 30 devices, which include both physical and emulated IoT devices, and general-purpose devices such as laptops, desktops, servers, etc., that are usually present in a Smart Lab. The dataset contains normal network traffic generated from various applications and attack network traffic comprised of 16 attack types, including IoT-based attacks, which are categorized into six categories. From the testbed, we collected 22.8 GB of raw network data and extracted 15.8 GB of preprocessed network data by using our own feature extraction approach. The preprocessed network data contains 121 flow-based features and 3 class labels of more than 34.8 million records. Additionally, this dataset covers extensive IoT-specific application protocols and a variety of normal network behaviours, which is a notable advantage over existing datasets. The MU-IoT dataset can be utilized to develop and validate machine learning-based Intrusion Detection and Mitigation Systems (IDMS). Moreover, the MU-IoT dataset helps to validate the centralized and federated learning-based IDMS. In addition, we also include the prior exploratory data analysis with the ranking of the features given by various feature selection algorithms as well as from our own feature ranking approach and the performance given by some common machine learning algorithms, as a future reference to the research community. The MU-IoT dataset is publicly available in the link https://manipuruniv.ac.in/CSDmuIOTdataset/.