HCCD-DS v2.0: A Transparent Synthetic Benchmark Dataset for Human–AI Decision Support Under Contextual Uncertainty

HCCD-DS v2.0 is a transparent and configurable synthetic benchmark dataset for studying human–AI decision support under contextual uncertainty. This release extends the original HCCD-DS benchmark by incorporating continuous user expertise, sequential trust adaptation, individual risk tolerance, explanation-sensitive decision behavior, and domain-specific profiles for healthcare, cybersecurity, and Internet of Things (IoT) scenarios.

The dataset contains 15,000 synthetic decision instances split into training and test subsets. Each instance links contextual risk, data completeness, system confidence, explanation metadata, simulated human accept/override behavior, override rationale, and final decision outcome. The revised generation framework is designed to support reproducible research on trust calibration, explanation-aware interaction policies, human intervention timing, and learn-to-defer algorithms.

This v2.0 release includes the regenerated dataset files, data generation scripts, analysis scripts, machine learning benchmark code, updated figures, benchmark results, and a changelog documenting the changes from the original release. The benchmark is fully synthetic and should not be interpreted as a replacement for real human–AI interaction logs. Instead, it is intended as a controlled, transparent, and reusable testbed for evaluating interaction-aware decision support methods under explicitly stated behavioral assumptions.