Prediction of ultimate strength for high strength concrete (HSC) using machine learning approaches - optimized by PSO technique

Abstract High strength concrete (HSC) is undoubtedly the most advanced building materials available nowadays. Its production involves simple steps with a variety of additives, including cement, water, fine and coarse aggregates, fly ash (FA), and ground granulated blast furnace slag (GGBFS). Although the interactions between these materials do not strictly follow a mathematical formula, the amounts of these ingredients show a major impact on the compressive strength. The most often used mechanical property for quality monitoring in concrete is its compressive strength after 28 days. It is crucial to have a tool that can directly simulate these interactions prior to production and casting the specimen. Machine learning (ML) models have shown to remain an effective technique to predict the concrete compressive strength, yielding results that can be more reliable than conventional. For the experimental data, the XGBoost Regression (XGB) model is the most dependable, with a $$R^2$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msup> <mml:mi>R</mml:mi> <mml:mn>2</mml:mn> </mml:msup> </mml:math> equal to 0.92, mean absolute error (MAE), and root mean squared error (RMSE) values of 2.92 MPa and 4.45 MPa . In addition, a comparison of particle swarm optimization was used to improve the relationship between input parameters and concrete compressive strength (CS). The study emphasises the accuracy with which machine learning approaches, specifically the XGB, can estimate the CS in building materials is higher than other models. It further provides researchers a swift and more reliable way to evaluate the effects of materials along with other factors on CS, eliminating the requirement for lengthy and expensive trial experiments.