Machine Learning-Driven Discovery and Optimization of Eco-Friendly Construction Materials

Abstract

The construction sector accounts for a substantial portion of global carbon emissions and resource exploitation, necessitating the development of eco-friendly materials that balance performance and sustainability. Traditional experimental approaches for material innovation are time-intensive and expensive, often limiting the pace of sustainable breakthroughs. This research develops a machine learning (ML)-based framework for identifying and optimizing eco-friendly construction materials by correlating compositional attributes and processing parameters with performance and environmental outcomes. Supervised learning models — including random forest regression, gradient boosting machines, and deep neural networks were trained on an extensive dataset compiled from peer-reviewed studies, encompassing over 600 material formulations. Target outcomes included compressive strength and embodied carbon indices, while predictors consisted of mix design variables and curing conditions. Gradient boosting demonstrated superior predictive performance (R² > 0.91), while feature analysis revealed the influence of supplementary cementitious materials and water/binder ratio on both performance and sustainability. Case evaluations illustrated optimized mixtures with comparable mechanical properties to conventional concrete but with 20– 35% lower environmental impact. This work highlights machine learning’s potential to streamline sustainable material design and reduce reliance on iterative physical experimentation. Future efforts will integrate multi-objective optimization and extend to broader classes of bio-based and recycled materials.