Machine learning prediction of compressive strength of hemp lime concrete for sustainable and resilient infrastructure

Abstract

The growing demand for sustainable and resilient infrastructure has increased interest in hemp–lime concrete as a low-carbon, bio-based construction material. However, its inherent heterogeneity makes accurate prediction of mechanical performance, particularly compressive strength, challenging. This study applies machine learning techniques to predict the 28-day compressive strength of hemp–lime concrete using Neural Networks and Random Forest models. A comprehensive dataset combining experimental and literature data is developed, incorporating formulation parameters, hemp properties, and specimen geometry. Model performance is evaluated using k-fold cross-validation. The Random Forest model achieves superior accuracy with R2 ≈ 0.90, while feature importance analysis identifies the key parameters influencing compressive strength. The proposed data-driven framework supports reliable performance prediction and mix design optimization for sustainable infrastructure applications.