Evaluating Parameter Influences on Planted Concrete Properties via Explainable Machine Learning Models

To investigate the complex functional relationships between pH, effective porosity, and compressive strength of planted concrete and their corresponding mixing ratios, a comprehensive database was developed from the relevant published literature. In this study, four machine learning (ML) algorithms were employed: a single algorithm—Multi-Layer Perceptron (MLP), and three ensemble algorithms—Gradient Boosting Regression (GBR), Extreme Gradient Boosting (XGBoost), and Random Forest Regression (RFR)—to predict the pH, effective porosity, and compressive strength of planted concrete. Additionally, the interpretable algorithm Shapley Additive Explanations (SHAP) was used to evaluate both global and local interpretations independent of the ML algorithms, providing insight into the decision-making process. The results demonstrate that the RFR algorithm achieved the highest R2 values of 0.93 (pH), 0.97 (effective porosity), and 0.94 (compressive strength) in predicting planted concrete properties, demonstrating optimal predictive performance. Furthermore, cement content was identified as the most influential factor affecting pH, while design porosity and maximum coarse aggregate size were the primary factors influencing effective porosity, in that order. For compressive strength, the two most critical factors were the water reducer and cement content.