Predicting lightweight concrete strength using stacking model and shap analysis

Abstract

This study proposes a stacking ensemble machine learning (ML) model for predicting the compressive strength of lightweight aggregate concrete. The model was developed and evaluated using a dataset comprising material composition parameters and experimentalcompressive strength results. The Stacking Ensemble model achieved high predictive performance with a correlation R of 0.906 and RMSE of 7.782 MPa on the testing dataset. Furthermore, the study employed SHapley Additive exPlanations (SHAP) to analyze the importance of input parameters on compressive strength. The SHAP analysis revealed that natural coarse aggregate content, water-to-binder ratio, natural lightweight aggregate content, and water content were the most influential factors. Two-dimensional SHAP analysis also elucidated the complex relationship between these parameters, indicating that the combination of low CLWA and W enhances the compressive strength. This research not only provides an effective prediction model but also opens up new avenues for optimizing lightweight aggregate concrete compositions based on SHAP analysis, thus contributing to improved quality and efficiency in the construction industry.