Meta: Toward a Unified, Multimodal Dataset for Network Intrusion Detection Systems

The lack of standardization across publicly available network intrusion detection datasets presents a significant challenge in developing generalizable machine learning-based models. These existing datasets often exhibit inconsistencies in feature sets and typically focus only on flow-level data, overlooking critical elements such as payload information and time-window-based contextual features. Such limitations make it difficult to detect sophisticated, time-sensitive attacks that rely heavily on both payload analysis and temporal patterns in network behavior. To address these challenges, we propose a unified multimodal dataset that integrates flow, payload, and contextual features from several renowned datasets. Our methodology implements a three-stage pipeline that processes raw packet capture (PCAP) files, extracts detailed metadata, and synchronizes the flow and payload data with time-based contextual features to ensure a comprehensive and enriched dataset. This approach allows for extensive cross-dataset validation, enabling the development of more robust and adaptable machine learning models for network intrusion detection. By providing a unified feature space and incorporating payload and contextual data, the dataset enhances the ability to detect a wider range of network threats with greater precision.

IEEE SOCIETY/COUNCIL Communications Society (COMSOC)

DATA TYPE/LOCATION CSV, Network traffic (flow and payload content); n/a

DATA DOI/PID 10.21227/d8at-gb29