IoT network traffic dataset using the custom flow representation

This dataset provides Custom Flow representations derived from
raw IoT network traffic traces, capturing detailed behavioral
characteristics of IoT communications. Each Custom Flow encapsulates
network behavior in a structured, vectorized format that includes
flow-level metadata, packet sequence timing, direction, and selected
payloads. Flows are uniquely identified by a five-tuple: device IP
address, remote IP address, protocol, device port, and remote port, and
maintain a fixed one-minute lifetime. To ensure consistent
temporal granularity and computational efficiency, long-lived connections
(such as persistent IoT–cloud sessions) are segmented into consecutive
flow records sharing the same identifier. The dataset was generated from
60 days of packet capture (PCAP) traces obtained from the publicly
available UNSW IoT Traffic Analytics platform. Two variants are included:
(1) Bidirectional Custom Flows, capturing both upstream and downstream
packets (~6 million flows), and (2) Unidirectional Custom Flows, capturing
only upstream packets from the device perspective (about 3.5 million
flows). Each day’s data is provided as a separate Parquet file, organized
and compressed by direction to facilitate scalable analysis. This dataset
provides a fine-grained yet computationally efficient representation of
Iot network behavior, supporting research in traffic analysis, anomaly
detection, and IoT device identification.