"Dataset and Code for Hybrid Knowledge Representation and Reasoning for Identifying Human Unsafe Behavior Underground"

"The data and codes include (1) The OWL 2 DL ontology file, along with the Python builder script for regenerating or extending the ontology; (2) the main evaluation framework implementing the rule base (10 SWRL-equivalent rules), feature extraction (37-dimensional feature vector), five ML classifiers, adaptive fusion algorithm, stratified 5-fold cross-validation, and noise robustness analysis (Section~\\ref{sec:noise_robustness}); (3) the per-scenario-type analysis script that produces per-scenario F1 scores and confusion matrices (Tables~14--16) using out-of-fold predictions, eliminating train-test leakage; (4) both balanced and unbalanced datasets (2,000 scenarios each) in JSON format; (5) a \\texttt{requirements.txt} for dependency management; (6) a README with execution instructions; and (7) rule base documentation with all the detection rules and their regulatory citations. All materials are released under the MIT License."