Explainable machine learning predictive model for mechanical strength of recycled ceramic tile-based concrete

Valorizing industrial byproducts in construction applications is a promising approach for enhancing sustainability. Global annual production of ceramic waste including broken tiles is a considerable challenge. Yet, such ceramic tile waste has great potential in sustainable concrete production, for instance as fine and coarse aggregate. Effective use of ceramic tile waste in concrete requires accurate prediction of recycled tile concrete mechanical strength. This study deploys state-of-the-art machine learning techniques for predicting the compressive and tensile strength of ceramic tile-based concrete. The authors recently performed an extensive experimental program on the mechanical characterization of ceramic tile-based concrete, allowing to build a comprehensive database of 252 data points with varying key mixture proportions. The latter was used to develop a data-driven machine learning (ML) modeling framework for predicting the mechanical properties of the concrete using XGBoost, CatBoost, LightGBM, and Extra Trees regressors. The independent variables of the dataset affecting mechanical strength included the cast density, percentage of ceramic tiles used as coarse and fine aggregate, water-to-cement ratio, water absorption capacity, and hydration age. The influence of different input features on the model predictions was visualized using SHAP feature importance plots. Ultimately, a machine learning-based and user-friendly graphical interface was created and made available through the Streamlit platform to aid in the design of ceramic tile-based sustainable concrete.