An autogluon-enabled robust machine learning model for concrete tensile and compressive strength forecast

Accurately predicting the compressive and tensile strength of concrete is crucial for the safety and dependability of infrastructure design and construction. Traditional empirical models, however, often fall short in prediction accuracy due to the intricate, nonlinear relationship between concrete strength and properties. This study introduces an innovative AutoGluon-Shapley additive explanations approach, which automates compressive and tensile strength prediction while providing meaningful result interpretations. Using the ConcreteXAI dataset of 3460 samples, electrical resistivity and ultrasonic pulse velocity were selected as input parameters for the models. 80% of the dataset was used for training while 20% was allocated for testing the AutoGluon-enabled models. The results demonstrate that the optimal compressive and tensile strength prediction models achieved root mean square errors of 1.38 and 0.11 MPa, respectively, in the testing phase. The analysis of input parameters revealed that ultrasonic pulse velocity was consistently a more reliable predictor for strength than electrical resistivity. The study concludes with a model that accurately predicts concrete compressive and tensile strength. The research limitation includes the use of only two input variables and the forecast of only compressive and tensile strengths. Future research may incorporate additional input and output parameters, including thermal conductivity and durability.