Enhancing concrete strength prediction models with advanced machine-learning regressors

Abstract

In recent years, widespread availability of large datasets has propelled the application of artificial intelligence and machine learning (ML) across various engineering domains. The study focuses on developing an optimised ML model to predict the strength of normal and high-strength concretes, utilising a comprehensive dataset comprising around 900 concrete mixes. Regression tools are employed to analyse the dataset and identify the most suitable regressor for accurate strength prediction. Performance indicators are then utilised to optimise the ML model on both training and test datasets. Results indicate that the random forest (RF) regressor, along with the XG Boost regressor, demonstrate strong correlation with the trained dataset. Validation based on experimental results from 50 laboratory concrete mixtures highlights the superiority of RF regressor's with low mean percentage error and superior standard deviation compared to other ML models.