Databases for synergistic multi-performance optimization of recycled aggregate concrete using explainable machine learning and NSGA-III
Description
This dataset comprises two literature-derived databases of recycled aggregate concrete compiled from publicly available academic literature for machine-learning prediction of 28-day compressive strength and slump. These databases underpin an associated study on synergistic multi-performance optimization using explainable machine learning and NSGA-III. The compressive-strength database contains 1,736 records and 15 columns, with 28-day compressive strength as the target variable, whereas the slump database contains 1,032 records and 15 columns, with measured slump as the target variable. Potential outliers were identified separately in the two databases using the Isolation Forest algorithm. All original records are retained; in the outlier_flag column, 0 denotes records retained for modeling and 1 denotes records excluded as potential outliers. All data were manually extracted from the source publications and harmonized to ensure consistent units and variable definitions. Detailed variable definitions, data-harmonization procedures, and source references are provided in the accompanying README.md file.
Files
Steps to reproduce
Open the two XLSX files; each data row represents one concrete mixture record. The source_reference column links each record to the corresponding numbered publication listed in the accompanying README.md file. Only 28-day compressive-strength results were included in the compressive-strength database, while zero-slump values were retained as valid observations in the slump database. Potential outliers were identified separately in the two databases using the Isolation Forest algorithm and recorded in the outlier_flag column. To reconstruct the datasets used for machine-learning analysis, retain records with outlier_flag = 0, yielding 1,649 compressive-strength records and 980 slump records. Use the 11 mixture-proportion and recycled-aggregate-property variables as model inputs and the 28-day compressive strength or slump as the corresponding prediction target. Detailed variable definitions, data-harmonization procedures, and source references are provided in the README.md file.
Institutions
- Guilin University of TechnologyGuangxi, Guilin