Dataset for anomaly detection in a production wireless mesh community network.

Area of Science:

• Computer Science
• Network Engineering
• Data Science

Background:

• Wireless Community Networks (WCNs) are growing globally, utilizing low-cost WiFi devices for user-built infrastructure.
• The dynamic nature of WCNs, with frequent node changes and heterogeneous links, poses significant challenges for anomaly detection.
• Existing research lacks comprehensive datasets for studying fault detection in these complex WCN environments.

Purpose of the Study:

• To present a novel dataset collected from a production WCN.
• To facilitate the investigation of unsupervised machine learning algorithms for fault detection in WCNs.
• To provide a benchmark for evaluating anomaly detection techniques in dynamic and heterogeneous network conditions.

Main Methods:

• Collected data from a central server aggregating information from 63 mesh nodes in a production WCN.
• Gathered diverse features including traffic, CPU, memory, and network topology (adjacency matrix, routing table, metrics).
• Included a known, unprovoked gateway failure event within the dataset for fault detection analysis.

Main Results:

• A comprehensive dataset from a real-world WCN is now available.
• The dataset contains rich information on network behavior, node performance, and topology.
• The inclusion of a specific fault event allows for targeted algorithm evaluation.

Conclusions:

• This dataset is the first of its kind for investigating fault detection in production WCNs.
• It provides a valuable resource for advancing research in WCN anomaly detection and machine learning applications.
• Enables empirical evaluation of algorithms designed for dynamic and fault-prone network infrastructures.