DB² Device Behaviour Fingerprinting Dataset Engineered Rolling-Window Features (v1.0) Overview This dataset provides engineered rolling-window behavioural features used for device fingerprinting and continuous device authentication research in networked and edge environments. Measurements were collected from 9 Raspberry Pi devices, across 4 CPU cores and 10 reboot sessions. Data Splits The dataset is partitioned to support strict, leak-free evaluation: FIT: Reboots 1–6 CAL: Reboots 7–8 TEST: Reboots 9–10 Identifiers and Privacy Device identifiers are anonymised using synthetic labels (D01–D09).Original hardware addresses have been removed and are not recoverable from this dataset. Raw traces are not included in this release. Features Features are engineered using rolling-window statistical aggregation over window sizes between 10 and 200 samples. Extracted statistics include: Mean Minimum Maximum Median Sum Standard deviation Skewness Kurtosis Quantiles (q10, q25, q75, q90) Interquartile range (IQR) Temperature residualisation and FIT-based centering (v3_centered_safe) have been applied. Device Identifier Alignment with Manuscript The manuscript refers to devices using identifiers D1–D9.This dataset uses zero-padded identifiers (D01–D09) to ensure stable lexicographic ordering. The correspondence between manuscript and dataset identifiers is: Manuscript ID Dataset ID D1 D01 D2 D04 D3 D06 D4 D08 D5 D03 D6 D02 D7 D05 D8 D07 D9 D09 This alignment is nominal only and does not affect data partitions, feature engineering, or evaluation procedures. Integrity Verification SHA-256 hashes for all CSV files are provided in SHA256SUMS.txt. Example (Windows): certutil -hashfile windows_FIT_corr_v3_PUBLIC_DXX.csv SHA256 Funding This work was supported by British Telecommunications plc (BT). License Creative Commons Attribution 4.0 International (CC BY 4.0). Citation Please cite the Zenodo DOI assigned to this record.