A Systematic Review of Machine Learning Techniques for Predicting Compressive and Flexural Strength of Mortars

Mortars are among the most used materials in civil construction, and their mechanical properties are directly related to several parameters, namely structural behavior, durability, and failure mechanisms. As a result, predicting these parameters is essential for preventive and corrective diagnosis of structures composed of this material. Recently, machine learning models have been applied to improve these predictions. Until recently, such predictions were based on traditional empirical procedures, which are unable to capture the variability of material parameters. Nevertheless, as this is still an emerging research topic, many limitations and gaps remain to be investigated, such as the dependence on data structure and the appropriate selection of hyperparameters. Furthermore, most studies applying machine learning to mortars focus on additions, admixtures, or durability- and performance-related aspects, rather than on assessing the efficiency and robustness of the models themselves. In this context, this systematic and critical review aims to identify how machine learning models have been applied to predict the mechanical properties of different mortars. The main input parameters used for this prediction are discussed, and the hyperparameters of several machine learning models are critically analyzed and discussed. The results indicate that tree-based ensemble models perform better than other approaches, with XGBoost showing superior predictive performance.