AT&DF-IDE Behavioural Device Fingerprint Dataset for Adversarial Attack and Defence Research

Le résumé fourni par la source

This dataset contains anonymised behavioural execution-response measurements collected from multiple Raspberry Pi single-board computers under controlled workload conditions. The data captures low-level system behaviour influenced by processor activity, timing dynamics, and hardware-dependent execution variability. These signals are used to create behavioural device fingerprints for identity-establishment experiments in continuous authentication environments. The release includes both raw execution-response observations and an engineered feature representation derived through systematic statistical aggregation, temporal analysis, and behavioural transformation procedures. The engineered features are designed to support advanced research tasks, including evaluating adversarial robustness, modelling attacks, validating defences, and analysing behavioural identity within distributed device ecosystems. To protect device privacy, all direct hardware identifiers have been removed. Devices are represented using stable anonymised identifiers, while device model information is maintained only at a general model-family level. This approach preserves analytical relevance while preventing exposure of sensitive deployment details. Internal experimental references and traceability artefacts have been excluded from the public release. The feature engineering strategy, processing pipeline, and research objectives in the AT&DF-IDE study differ in scope and design. Therefore, this dataset is intended as a processed behavioural fingerprint resource specifically supporting adversarial robustness research in identity-establishment mechanisms. This release aims to facilitate reproducible research in behavioural device authentication, security evaluation of fingerprint-based identity mechanisms, and analysis of attack-resilient device identification methods.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.