A Fairness-Constrained Risk-Adaptive Access Control Method

This dataset contains the experimental data used for the paper “A Fairness-Constrained Risk-Adaptive Access Control Method”. The released files are prepared to support reproducibility of the proposed safety- and fairness-constrained risk-adaptive access control experiments. The dataset covers three access-control-related scenarios: CASAS Smart Home: smart-home activity records mapped into access request samples for behavior-based risk assessment. UNSW-NB15: network traffic records mapped into access request samples for network access risk identification. LANL Authentication Logs: enterprise authentication records sampled and mapped into access request samples for high-risk authentication behavior detection. For each dataset, the original records were transformed into a unified access-control representation. Each access request is represented using risk-related indicators, including privilege level, historical risk, temporal activity deviation, source–destination anomaly, operation compliance, abnormal request ratio, and interaction entropy. The risk labels are defined as low, medium, and high. For fairness evaluation, the high-risk class is further converted into a binary task, where high is treated as the security-sensitive positive class and low/medium are merged as non-high. The released data may include: mapped access request files; training, validation, and test splits; label and group attribute files; dataset statistics; preprocessing and mapping metadata; sampling reports for the LANL authentication subset. The protected group attribute is used only for fairness evaluation and post-processing. It is not used as an input feature of the base risk prediction model. This dataset is intended for research on: risk-adaptive access control; high-risk access request identification; fairness-aware security decision-making; safety-constrained fairness post-processing; performance–security–fairness trade-off analysis. Please note that this release is derived from publicly available benchmark datasets or their processed experimental subsets. Users should cite the original data sources where applicable, including CASAS Smart Home, UNSW-NB15, and LANL authentication datasets, as well as the associated paper that describes the experimental construction.