Robust Machine Learning Framework for Modeling the Compressive Strength of SFRC: Database Compilation, Predictive...

Area of Science:

• Construction Engineering
• Materials Science
• Data Science

Background:

• Construction engineering increasingly integrates machine learning (ML) for material property forecasting.
• Existing models for steel-fiber-reinforced concrete (SFRC) often lack transparency and practical applicability.
• There is a need for advanced, interpretable ML models to predict SFRC characteristics.

Purpose of the Study:

• To develop and validate an accurate machine learning model for predicting the compressive strength of SFRC.
• To identify optimal SFRC mix designs using advanced ML techniques.
• To bridge the gap between theoretical ML models and practical applications in construction.

Main Methods:

• Utilized the extra gradient (XG) boosting algorithm for predictive modeling.
• Compiled a comprehensive database from 43 publications (420 records) focusing on crimped, hooked, and mil-cut fibers.
• Conducted experimental validation with 20 SFRC mixtures and employed partial dependence plots (PDPs) for analysis.

Main Results:

• The XG boosting model achieved high accuracy, with a mean target-prediction ratio of 99% on independent datasets.
• Identified optimal SFRC formulations with enhanced compressive strength.
• Partial dependence plots revealed key relationships between input parameters and SFRC compressive strength.

Conclusions:

• The developed ML model offers a transparent and accurate tool for predicting SFRC compressive strength.
• The study provides practical insights for optimizing SFRC mix designs in construction.
• A user-friendly digital interface was created to facilitate professional adoption and application.