An intelligent selection method for mix proportion of engineered cementitious composites

Engineered Cementitious Composite (ECC) has attracted significant attention due to its exceptional tensile ductility and crack control capabilities. However, its high material cost limits its widespread engineering application. To achieve intelligent and cost-effective mix design, this study introduces for the first time an integrated machine learning framework based on a dataset of over 300 detailed mix proportions, and systematically evaluates the performance of 12 machine learning algorithms in predicting key mechanical properties of ECC, including compressive strength, tensile strength, and tensile strain.The results indicate that the CatBoost model demonstrates the highest prediction accuracy and generalization capability, with a comprehensive performance evaluation index ( P -value) of 0.434. Feature importance analysis and SHAP value interpretation consistently reveal that the water-to-binder ratio is the most critical factor influencing all mechanical properties, followed by silica fume content and cement dosage. Furthermore, the elongation at break of fibers also shows a significant impact on performance. To establish a closed-loop process from prediction to design, this study proposes an innovative quantitative optimization method. Candidate mix proportions are generated within a target compressive strength range, and a newly defined Comprehensive Performance-Benefit Index ( CP ) is used to balance tensile performance and economic cost. A case study validates the effectiveness of this approach and identifies the mix proportion with the optimal overall benefit ( CP = 0.834) from multiple candidates. This study demonstrates that the proposed data-driven framework can efficiently and accurately guide the mix design of ECC.