An Industrial Cokemaking Coal-Blending and Coke-Quality Dataset

Each record aggregates the individual coals in one blending list into 65 blend-level descriptors: mass-weighted coal-quality means (proximate indices Ad, Vdaf, Std, G; vitrinite reflectivity and active-component terms; Gieseler-fluidity characteristic temperatures; dilatometer terms), within-blend coefficients of variation, measured feed-coal properties (Mt, Ad, Vdaf, Std, G, particle-size fractions, coking temperatures), engineered interaction and polynomial terms, per-coal-type mass fractions classified by GB/T 5751, and blend cost — together with 7 measured coke-quality responses (Ad, Vdaf, Std, M10, M40, CRI, CSR).

A strict temporal train/validation/test split (160/23/46, random seed 42) is provided in the `split` column. The released table is complete (no missing cells). Advanced coal-quality indicators, missing in 11.1% of individual-coal records, were imputed using training-only category and global means before list-level aggregation; plastic-layer X/Y indices and ash composition were not measured and are therefore absent.