Intelligent data-driven compressive strength prediction and optimization of reactive powder concrete using multiple ensemble-based machine learning approach

Abstract

In recent years reactive powder concrete (RPC), also known as ultrahigh-performance concrete, emerged as one of the most efficient building materials due to its ultrahigh compressive strength (usually greater than 150 MPa), ductility, and durability. The existing literature presents a wealth of experimental data exploring the intricate interactions among the various components within RPC mixtures, emphasizing the requirement for a potent and efficient model to optimize, simulate, and enrich the understanding of RPC mixture characteristics. This study presents a notable attribute utilizing a multilayer stacked model that harnesses the potential to harmoniously seamlessly integrate and synergize various machine learning (ML) algorithms. An extensive dataset from reliable resources in the literature was used to develop a new ML model for predicting the compressive strength of RPC. It was found that there is a strong positive correlation between the RPC’s compressive strength and the dosage of silica fume, fiber dosage, and age of concrete. In contrast, water content and dosage of GGBFS were observed to have a low correlation with the compressive strength of the RPC. The most significant features were age, fiber dosage, silica fume content, cement content, HRWR dosage, fine aggregate content, and water content. Further, in this study, predictive efficacy identifies the stacking algorithm as optimal, followed by XGBoost, RF|ETR, and KNN. Notably, R2 strengths vary: stacking achieves 0.96, XGB sustains 0.954, RF and ETR hold 0.95, while KNN reaches 0.77.