Use of soft computing approaches for the prediction of compressive strength in concrete blends with eggshell powder

Abstract

The present study showcases a prediction model for estimating the compressive strength of concrete combined with ESP utilizing machine-learning techniques. The models were created using 399 datasets that were sourced from published literature. The datasets included a range of input factors, including cement content, ESP content, fine aggregate content, coarse aggregate content, water content, and curing duration. The models used the compressive strength of ESP mixed concrete as the output variable. This research used a collection of seven machine learning algorithms, namely linear regression, artificial neural network, boosted decision tree regression, K nearest neighbors, random forest regression, support vector regression, and XGboost as statistical evaluation tools in order to determine the best accurate and dependable model for forecasting the compressive strength of ESP mixed concrete. Among the machine-learning models assessed in this investigation, the XGboost model has shown exceptional efficacy in forecasting compressive strength. It attained an R2 value of 0.99 and an RMSE of 0.99 MPa for the training dataset while reaching an R2 value of 0.82 and an RMSE of 4.48 MPa for the testing dataset. The sensitivity analysis results of the XGboost model indicate that the compressive strength of the material is mostly affected by the curing period. The compressive strength of a material is also significantly impacted by the amounts of cement and water in the mix.