MACHINE LEARNING APPLICATIONS IN SUSTAINABLE CONCRETE MIX DESIGN OPTIMIZATION: A COMPREHENSIVE REVIEW OF PERFORMANCE, DURABILITY, AND ENVIRONMENTAL BENEFITS

Abstract

Concrete production accounts for approximately 8% of global CO₂ emissions, necessitating urgent adoption of sustainable practices through innovative design methodologies. Machine learning (ML) has emerged as a transformative technology enabling rapid optimization of concrete mix designs that balance mechanical performance, durability, and environmental sustainability. This comprehensive review synthesizes recent advances in ML applications for concrete mix design, examining prediction accuracy, multi-objective optimization frameworks, and integration with supplementary cementitious materials (SCMs). A systematic analysis of 47 peer-reviewed studies from 2016–2026 reveals that ensemble learning algorithms (XGBoost, Random Forest, Gradient Boosting) and deep neural networks consistently achieve prediction coefficients of determination (R²) exceeding 0.90, with mean absolute errors below 5%. Multi-objective optimization approaches have successfully reduced CO₂ emissions by 25–51% while maintaining target strengths of 30–70 MPa. Integration of interpretability tools (SHAP, LIME) demonstrates that water-to-binder ratio, cement content, and supplementary material dosage are dominant performance drivers. This review identifies critical research gaps including long-term durability validation of ML-optimized mixes, real-time field implementation protocols, and standardization of datasets across geographic regions. Future directions emphasize hybrid physics-informed neural networks, automated quality assurance systems, and circular economy applications utilizing construction-demolition waste. ML-driven concrete optimization represents a viable pathway toward achieving sustainable infrastructure goals while reducing material costs by 15–30% and embodied carbon by up to 51%.