Experimental data and machine-learning code for composition–strength relationships in electrolytic manganese residue–municipal solid waste incineration fly ash binders

Description

This dataset provides experimental data, statistical analyses, and machine-learning code supporting the study entitled “Composition–strength relationships and interpretable modelling of electrolytic manganese residue–municipal solid waste incineration fly ash binders”. The experimental investigation comprises 45 multi-waste cementitious binder formulations designed using a 3 × 3 × 5 full-factorial approach. The materials include thermally treated electrolytic manganese residue (REMR), coal fly ash (FA), and municipal solid waste incineration fly ash (MSWI FA). The accompanying supplementary tables (S1–S34) contain mixture proportions, fresh-state properties, compressive strength results, statistical analyses, physicochemical characterization, environmental leaching results, and machine-learning evaluation metrics. The machine-learning dataset comprises 180 mixture–age records, including 174 quantitative strength records used for modelling and six non-quantifiable (NQ) records excluded from model training and evaluation. The computational files include Python scripts, model inputs, grouped cross-validation procedures, feature analysis, model evaluation results, and supporting documentation. The dataset is intended to facilitate verification of the reported experimental findings and reproducibility of the machine-learning analyses. The accompanying documentation describes the organization of the files, variable definitions, data-processing procedures, and computational workflow.

Steps to reproduce

1. Download the supplementary Excel workbook and the machine-learning reproducibility package from this dataset. 2. Open the supplementary workbook to access the experimental results and statistical analyses associated with Tables S1–S34, including mixture proportions, fresh-state properties, compressive strength, material characterization, and leaching data. 3. Extract the machine-learning ZIP archive and consult the included README and environment specification for the required Python dependencies, file organization, and execution instructions. 4. The machine-learning records cover 45 binder formulations and 180 mixture–age observations. Of these, 174 contain quantitative compressive-strength measurements used for modelling, while six non-quantifiable (NQ) records are excluded from model training and evaluation. Any zero values used as placeholders for NQ records must not be interpreted as measured strengths. 5. Follow the documented scripts and execution sequence to reproduce data preprocessing, mixture-grouped cross-validation, model evaluation, and interpretability analyses. 6. Compare the resulting numerical outputs with the corresponding supplementary tables and reported manuscript results. Consult the README for variable definitions, model configurations, and any environment-specific requirements.

Institutions

• Guizhou University

  Guizhou, Guiyang
• Guizhou Institute of Technology

  Guizhou, Guiyang

Categories

Licence