Development of machine learning models for the prediction of the compressive strength of calcium-based geopolymers

Compressive strength is an important mechanical index that determines the mixture design of geopolymer, and its accurate prediction is essential. The existing experiment-based and statistical methods are time-consuming, labor-intensive and inaccurate. This study aims to develop an effective, reliable and interpretable machine learning (ML) model for predicting the compressive strength of calcium-based geopolymers. Feature engineering was constructed with molar ratios of raw material oxide composition, curing system, and mixing design. A total of eight algorithms in three types, traditional ML algorithms, integrated tree-based ML algorithms, and deep neural network algorithm, were employed to predict the compressive strength, and their differences, advantages, and disadvantages were compared. The importance of input variables in model training was evaluated. The contribution and influence pattern of input features on the development of compressive strength were revealed using the SHapley Additive exPlanations (SHAP) and inverse prediction. The results demonstrate that among the eight models proposed, the XGB model had the highest prediction accuracy (91%) and the lowest root mean squared error (3.85 MPa). Based on the importance analysis and the SHAP value, the parameters that had the greatest impact on the compressive strength were curing age, n(H2O)/n(Na2O), curing temperature, n(SiO2)/n (CaO) and the mass ratio of alkali activation solution to solid powder (L/S). The effects of input features on the compressive strength development of calcium-based geopolymers captured by SHAP and inverse predictions based on the best predictive model were consistent with the experimental results and theoretical understanding. The research in this paper facilitates the rapid prediction, improvement and optimization of the proportioning design and application of calcium-based geopolymers, and also provides a theoretical basis for the utilization of industrial and construction waste, in line with sustainable and low-carbon development strategies.