Multi-target ML prediction of residual flexural strength in steel fibre-reinforced concrete

This article proposes a novel approach for constructing ground-supported structural slabs. It applies machine learning (ML) to predict the residual flexural strength of fibre-reinforced concrete under bending. Adaptive Boosting (AdaB) estimates post-peak flexural strength of steel-fibre-reinforced concrete at two crack-width levels. A few of these crack widths are 0.5 mm (fR,1) as well as 2.5 mm (fR,3). Two advanced optimisation methods, Catch Fish Optimization (CF) and Coati Optimization (Co), are assessed for this reason. The AdaB models’ performance is significantly influenced by their hyperparameters, which can be changed through optimisation methods. In order to obtain data, 216 test samples were collected. The dataset was split into 20% (44) for evaluation, and the rest 80% (172) was used in learning and model construction. Various pre-processing steps are carried out to introduce the dataset to the models. Tenfold cross-validation, the normalisation and randomisation process and feature importance are some of these steps. The results indicated that AdaB-CF has the smallest U95% values of 1.82496 and 1.4848 (fR,1) and 2.1401 and 1.709 (fR,3), but AdaB-Co displays higher U95% values during both the training and assessment stages, recording 1.9244 and 1.7138 (fR,1) and 2.3463 and 1.7951 (fR,3). All models are reliable, but AdaB-CF shows superior performance.