Estimating mechanical and dynamic properties of rubberized concrete using machine learning techniques: a comprehensive study

Abstract

Purpose Currently, many experimental studies on the properties and behavior of rubberized concrete are available in the literature. These findings have motivated scholars to propose models for estimating some properties of rubberized concrete using traditional and advanced techniques. However, with the advancement of computational techniques and new estimation models, selecting a model that best estimates concrete's property is becoming challenging. Design/methodology/approach In this study, over 1,000 different experimental findings were obtained from the literature and used to investigate the capabilities of ten different machine learning algorithms in modeling the hardened density, compressive, splitting tensile, and flexural strengths, static and dynamic moduli, and damping ratio of rubberized concrete through adopting three different prediction approaches with respect to the inputs of the model. Findings In general, the study's findings have shown that XGBoosting and FFBP models result in the best performances compared to other techniques. Originality/value Previous studies have focused on the compressive strength of rubberized concrete as the main parameter to be estimated and rarely went into other characteristics of the material. In this study, the capabilities of different machine learning algorithms in predicting the properties of rubberized concrete were investigated and compared. Additionally, most of the studies adopted the direct estimation approach in which the concrete constituent materials are used as inputs to the prediction model. In contrast, this study evaluates three different prediction approaches based on the input parameters used, referred to as direct, generalized, and nondestructive methods.