Prediction Models with GUI for Splitting Tensile Strength of Polypropylene-Fiber-Reinforced Recycled Aggregate Concrete

The construction industry’s growth trend has resulted in a considerable huge volume of demolished concrete. The proper usage of recycled aggregate (RA) obtained from demolished construction in concrete production provides a potential sustainable alternative to natural aggregates, but it leads to a loss in concrete strength. Incorporating polypropylene fibers (PPFs) into concrete with RA is considered an effective way to enhance the properties of RA concrete. However, the variability in RA and PPF properties makes correct estimates of splitting tensile strength (TS) extremely challenging. This study focuses on applying two machine learning (ML) algorithms, Extreme Gradient Boosting (XGB) and Random Forest (RF) improved by Particle Swarm Optimization (PSO), to predict the TS of polypropylene-fiber recycled-aggregate concrete (PPF-RA-concrete). To develop the models, a dataset of 556 TS data points was used with different amounts of twelve input parameters acquired from the literature. The input parameters covered the mixture proportions coupled with the properties of RA and PPF. The results show PSO-XGB achieved R2 = 0.9708, a root mean squared error of 0.22 MPa, a mean absolute error of 0.152 MPa and a mean absolute percentage error of 4.74%, outperforming PSO-RF. Shapley additive explanations, individual conditional expectation and partial dependence plots were used to examine model responses. The results revealed that the RA replacement ratio has the largest negative impact on TS prediction, while the curing age has the largest positive impact on TS prediction, followed by the aspect ratio of PPF. The created Graphical User Interface (GUI) acts as an excellent tool for estimating TS of PPF-RA-concrete within the range of the dataset used.