Data-Driven Analysis of Strength-Influencing Factors in Solid-Waste-Based Calcium Sulfoaluminate Cement

Calcium sulfoaluminate cement (CSA), which can be produced with a high proportion of industrial solid waste, presents a promising pathway for the transition of the building materials industry to low-carbon components. However, the complex and variable composition of such waste materials significantly influences the performance of CSA, thereby limiting its widespread application. To address this limitation, this study employed machine-learning (ML) techniques to perform a data-driven analysis of the factors affecting the strength of solid-waste-based CSA. Key parameters from raw material preparation to paste formation were investigated. Multiple ML models, including Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest (RF), Adaptive Boosting (AdaBoost), Gradient Boosting Decision Tree (GBDT), and eXtreme Gradient Boosting (XGBoost), were employed to predict the compressive strength of CSA. SHapley Additive exPlanations (SHAP) analysis was applied to interpret the contribution of each feature. The results show that ensemble models generally outperform traditional ones, with the XGBoost model demonstrating superior performance (R2 = 0.99 for training and 0.96 for testing) and prediction errors predominantly falling within ±5 MPa. Feature importance analysis identified curing age, Al/Fe ratio, and aluminum–silicon ratio (N) as the most influential factors, followed by calcination temperature, gypsum content, aluminum–sulfur ratio (P), and alkalinity coefficient (Cm). Extending the curing age, increasing N, and adding gypsum were found to enhance compressive strength, whereas increasing P, Cm, calcination time, and the water-to-cement ratio (W/C) negatively affected strength. Furthermore, sample clustering enabled the identification of optimal ranges for specific features. The data-driven and interpretable approach adopted in this study is expected to facilitate rapid evaluation of the compressive strength of solid-waste-based CSA and to support performance optimization strategies.