The Gotham Dataset is a large-scale, reproducible benchmark for evaluating decentralized Intrusion Detection Systems (IDS) and Federated Learning in virtualized smart cities. It captures interface-level network traffic from 78 heterogeneous IoT devices, including complex attack vectors like Mirai botnets, Merlin C2 traffic, and CoAP amplification, preserving the non-IID nature of edge data for realistic AI security training.
This Industrial IoT dataset provides synthetic yet realistic sensor data simulating equipment operation under normal and various failure conditions. Designed for predictive maintenance and machine learning, it includes sensor specifications, operational thresholds, and failure labels, allowing researchers to develop anomaly detection models without the constraints of sensitive real-world industrial data.
A multi-sensor dataset from a real manufacturing environment monitoring CNC machine degradation through vibration, temperature, current, and acoustic emission sensors over a 6-month period.
MQTT_UAD is a public MQTT traffic dataset published in Data in Brief 2025, containing labeled benign and attack scenarios in IoT networks that use the MQTT protocol, designed for training and evaluating intrusion detection systems.
DataSense is a real-time Industrial IoT (IIoT) dataset from the Canadian Institute for Cybersecurity, combining synchronized sensor and network data from a 40-device testbed with over 15 types of industrial sensors for anomaly and intrusion detection research.
An MQTT DoS and DDoS IoT attack dataset collected on a Raspberry Pi 3B+ Mosquitto broker over 12 sessions, including three days of normal traffic and several minutes of attack traffic, totaling 424,716 labeled entries for machine learning-based IDS and IPS research.
A real-world MQTT-based IoT cybersecurity dataset collected from the MQTTEEB testbed at the International University of Rabat, with benign traffic and five attack types (DoS, SlowITe, Malformed Data Injection, Brute Force, Publish Flooding), provided in multiple processed forms (raw, cleaned, normalized, standardized, SMOTE) for AI-driven intrusion detection research.
A synthetic but carefully constructed IoT dataset for smart home renewable energy management, providing five CSV datasets (20, 50, 100, 200 homes over 365 days) with daily energy consumption and production values to simulate small, medium, and large-scale smart city and smart home scenarios.
An image dataset of 3,200 manually annotated RGB images covering 32 classes of IoT education kits (Arduino, ESP32, Raspberry Pi, etc.), collected from multiple sources and annotated with polygon masks via Roboflow for object detection research in educational IoT kits.
A merged and optimized dataset combining N-BaIoT (IoT-specific traffic) and UNSW-NB15 (general network threats) with feature engineering, dimensionality reduction, and benchmarked ML models (Decision Tree, SVM, Random Forest, Neural Network) for IoT anomaly detection, published in Scientific Reports.
A federated learning evaluation across several contemporary IoT and IIoT intrusion detection datasets, benchmarking algorithms such as FedAvg, FedProx, and FedNova with LSTM and Transformer models in in-domain, cross-dataset, and multi-dataset federation scenarios.
IoT-DH is a real-world IoT DDoS honeypot dataset collected from a honeypot deployment and converted from PCAP to CSV with traffic features and labels for DDoS classification, identification, and detection tasks.