Compressive strength, flexural strength, and slump of recycled aggregate fibre-reinforced fly ash concrete using explainable extreme gradient boosting machine learning model with prediction tool

This study develops machine learning predictive models to evaluate the compressive strength ( f' c ), flexural tensile strength ( f r ), and slump of recycled aggregate fibre-reinforced fly ash concrete (RAFRC-FA). The models were developed using a database compiling 1028, 531, and 430 records for f' c , f r , and slump, respectively, with 21 input parameters related to concrete constituents, fly ash, fibres, and concrete testing age (for f' c and f r ). A total of 8 machine learning models representing single and ensemble algorithms were adopted in this study. The results revealed that the extreme gradient boosting (XGB) model outperformed all models, with mean absolute error and coefficient of determination of 2.10 MPa and 95.11% for f' c , 0.26 MPa and 92.16% for f r , and 15.54 mm and 85.98% for slump, respectively. Moreover, the XGB model exhibited the lowest standard deviation (0.024 and 0.042) and coefficient of variance (2.36% and 4.20%) of predicted-to-actual ratios compared to conventional analytical models for f' c and f r , respectively. In addition, the SHapley Additive exPlanation (SHAP) tool illustrated that concrete ingredients were the most influential factors affecting the compressive and flexural strength of RAFRC-FA, whereas the aggregate properties exhibited the highest impact on the slump of RAFRC-FA. Furthermore, a web-based application was developed and verified using unseen data for the prediction of the mechanical properties of RAFRC-FA. The experimental-to-predicted ratios of the predictions of the f' c , f r , and slump by the web-based application were in the range of 1.00 to 1.13, 0.99 to 1.06, and 1.00 to 1.13, respectively. • Mechanical properties of fibre-recycled aggregate concrete with fly ash were predicted using machine learning. • Properties tested were concrete compressive strength, flexural strength, and slump. • XGB model surpassed all ML models with an R 2 of 95.11% for compressive strength. • The XGB ML predictive model outperformed existing analytical models. • A web-based prediction tool was established based on the XGB model.