PowerTrace-IoT

PowerTrace-IoT is a multi-modal dataset capturing power consumption, environmental, and inertial telemetry from five ESP32-based IoT sensor nodes under normal operation and seven distinct physical and software attack classes. The dataset is designed to support research in machine-learning-based intrusion detection systems (IDS) leveraging physical-layer side-channels. To ensure strict out-of-distribution (OOD) evaluation and prevent temporal data leakage, the data was collected across two independent sessions: Training Corpus (12h): High-density attack injection schedule ensuring adequate representation of all attack vectors, including rare coordinated botnet events (101,422 readings). Evaluation Corpus (48h): Stochastic attack schedule with natural diurnal baseline variations, replicating realistic field deployments (430,679 readings). Included Data: Calibrated power telemetry (voltage, current, power, and variance computed via Welford’s algorithm on the edge). Environmental context (temperature, humidity). Inertial context (3-axis acceleration and angular velocity). Event-driven, sub-second ground-truth labels for 7 attack classes (DDoS, CPU Exhaustion, Kamikaze, Firmware Tampering, Sensor Spoofing [DHT11 & MPU6500], and Connection Loss). This archive includes the cleaned CSV files for both sessions, the raw hardware traces for transparency, database schemas, and all attack injection logs to ensure full reproducibility.