IoT Data Stream

Synthetic IoT Sensor Stream Dataset Overview This dataset contains a synthetic multivariate time-series stream generated from 3 independent IoT sensors. It is specifically designed for testing anomaly detection, time-series forecasting, and stream-poisoning mitigation algorithms. The data spans a total of 1,500,000 samples (500,000 per sensor) recorded at a high-frequency sampling rate of 100 Hz. It features a realistic stationary Gaussian noise background interspersed with artificially injected contiguous step-change anomalies. Features The dataset contains the following physical and network properties measured across all sensors: Temperature (`temp`): Measured in degrees Celsius (°C). Pressure (`pres`): Measured in kilopascals (kPa). Throughput (`thr`): Measured in kilobytes per second (KB/s). Sensor ID (`sensor_ids`): A string identifier indicating the source sensor (`sensor_0`, `sensor_1`, or `sensor_2`). Timestamp (`timestamps_ms`): Absolute timestamp in milliseconds. The base epoch begins at midnight `2024-01-01 UTC`, with a simulated 3ms transmission jitter offset interleaved per sensor. Target Variable (Ground Truth) Label (`y`): A binary label indicating the state of the system at that timestamp. `0` = NORMAL `1` = ANOMALY Data Characteristics Normal State (Background Noise) During normal operation, each sensor emits stationary Gaussian noise based on distinct nominal distributions: Sensor 0: Temp ~ N(22.0, 0.5²), Pres ~ N(101.3, 0.2²), Thr ~ N(512.0, 50.0²) Sensor 1: Temp ~ N(35.0, 1.0²), Pres ~ N(98.5, 0.4²), Thr ~ N(256.0, 30.0²) Sensor 2: Temp ~ N(18.5, 0.3²), Pres ~ N(102.1, 0.1²), Thr ~ N(1024.0, 80.0²) Anomalous State Anomalies account for approximately **5%** of the total dataset. Injection Method: Anomalies are injected as contiguous step-changes (blocks of ~1% length of a sensor's stream). Magnitude: During an anomalous event, the readings shift by **±5.0 standard deviations** from the nominal mean across all features for that specific sensor. File Structure The data is provided in a compressed NumPy format (`.npz`) and includes standard scaler parameters fitted strictly on the normal baseline data. `iot_stream.npz`: Contains the arrays `X` (scaled features), `X_raw` (unscaled features), `y` (labels), `sensor_ids`, and `timestamps_ms`. The arrays are pre-sorted chronologically to simulate a true interleaved multi-sensor data stream. `iot_stream_scaler.json`: Contains the mean and scale parameters used by `StandardScaler` to normalize the data.