Data Center Temperature IoT Dataset for Anomaly Detection (Zenodo 2025)

Overview

The Data Center Temperature IoT Dataset (DAD - Data center Anomaly Detection) published on Zenodo in January 2025 provides real-world temperature monitoring data from a physical data center for anomaly detection research.

Data Collection

• Temperature measurements from NFC (Near Field Communication) smart passive sensor technology deployed in a physical data center.
• Four refrigeration unit sensors connected to an internal IoT network, monitoring critical cooling infrastructure.
• Virtual infrastructure consisting of five virtual machines, one MQTT broker, and four client nodes for data collection and distribution.
• Time-series data sampled at regular intervals capturing both normal operational patterns and injected anomalies.

Data Characteristics

• Normal behavior: Typical temperature patterns during standard data center operations, including daily and weekly cycles.
• Anomalous behavior: Labeled anomalies reproducing real-world failure scenarios such as cooling system malfunctions, sensor drift, and environmental incidents.
• Mathematical modeling using time-series techniques to approximate realistic data center thermal dynamics.
• Network-level visibility: Data observed from the perspective of the IoT network, including MQTT message traffic and sensor communication patterns.

Dataset Format

• Time-stamped temperature readings from each sensor with device identifiers.
• Binary or multi-class anomaly labels indicating normal vs. anomalous periods.
• Metadata including sensor locations, refrigeration unit IDs, and anomaly descriptions.
• MQTT message logs capturing IoT protocol-level communication.

Use Cases

• Anomaly detection: Training and evaluating time-series anomaly detection algorithms for critical infrastructure monitoring.
• Predictive maintenance: Developing early warning systems for data center cooling system failures.
• IoT protocol analysis: Research on MQTT-based sensor networks and message pattern analysis.
• Edge computing: Benchmarking lightweight anomaly detection models suitable for deployment on edge devices in resource-constrained environments.