Experimental Results and Reproducibility Package for Neuro-Symbolic Validation of Cluster-Based Risk States in Gas Pipeline SCADA Systems

This record contains the experimental results and reproducibility materials for the study “A Neuro-Symbolic Validation Framework for Cluster-Based Cyber-Physical Risk States in Gas Pipeline SCADA Systems”. The package includes dataset characterization tables, leakage-clean preprocessing records, machine learning baseline results, neuro-symbolic validation results, cluster-based risk-state summaries, symbolic rule activation statistics, alpha sensitivity analysis, explanation examples, publication-ready figures, metadata files, and reproducibility scripts. The experiments were conducted on the New Gas Pipeline and Legacy Gas Pipeline SCADA datasets using their public ARFF versions. The New Gas Pipeline dataset was processed using the target attribute “categorized result”, where class 0 denotes normal behavior and non-zero classes denote cyber-physical risk conditions. The Legacy Gas Pipeline dataset was processed using the target attribute “result”, where class 0 denotes normal behavior and non-zero classes denote risk states. Target-related attributes were excluded from the input feature space to reduce target leakage. The package does not redistribute the original SCADA datasets. Dataset sources and preprocessing metadata are documented in the data source log. The purpose of this record is to support reproducibility of the reported experimental results, including machine learning baseline evaluation, cluster-based risk-state construction, Boolean symbolic validation, ML-symbolic conflict detection, and explanation coverage analysis.