A comprehensive study on estimating the primary crack spacing of flexural reinforced concrete components using machine learning techniques

Accurate estimation of primary crack spacing in flexural reinforced concrete components is essential for serviceability assessment and durability-related decision making. This study evaluates eleven regression-based machine learning models, combined with nine preprocessing strategies, to predict primary mean crack spacing using 96 experimental observations collected from the literature. A systematic hyperparameter search produced approximately 3500 trained model instances. Model performance was assessed using the correlation coefficient, normalized root mean square error, and normalized mean absolute error. Extremely randomized trees consistently provided the best generalization, achieving testing performance of R up to 0.97 with normalized root mean square error and normalized mean absolute error as low as 5% and 4%, respectively, depending on preprocessing (Original, Standardized, Normalized, and Forward Selection). Benchmarking against Eurocode 2, fib Model Code 2010, Reineck’s approach, and the strain compliance method showed that the extremely randomized trees-based models aligned most closely with the experimental crack spacing measurements within the range of parameters represented in the database. These results indicate that properly tuned tree-ensemble models can support serviceability assessment, rapid screening of design alternatives, and benchmarking of code-based crack-spacing checks within the parameter range represented by the compiled database.