Predictive analysis of concrete tensile strength using metaheuristic algorithms helping a neural network.

In most construction projects, concrete is a widely used material in different structural elements due to its suitable mechanical behavior. Therefore, it is crucial to achieve a dependable estimation of the concrete mechanical parameters such as tensile strength. In this research, a well-known machine learning model-multi-layer perceptron neural network (MLPNN)-is optimized by multi-tracker optimization algorithm (MTOA) to avoid computational insufficiencies. The model predicts the splitting tensile strength of concrete based on the features of the concrete mixture. For validation, MTOA is compared to multi-verse optimizer (MVO), crow search algorithm (CSA), and backtracking search algorithm (BSA). Primary results showed that all optimized ANNs can reliably understand and predict the tensile strength pattern. However, from accuracy comparison, MTOA-MLPNN > MVO-MLPNN > CSA-MLPNN > BSA-MLPNN, based on respective percentage errors of 10.75, 11.28, 13.70, and 14.58, as well as correlation coefficients of 0.92, 0.90, 0.87, and 0.83. Moreover, the computational cost of the MTOA-MLPNN was found to be lower than the mentioned benchmarks. In addition, this model was validated using 10-fold cross-validation, and also its accuracy surpassed several models from earlier literature. All in all, the hybrid of MTOA-MLPNN created a scientifically novel and efficient predictive framework for the low-cost analysis of concrete tensile strength and mixture design.