Development of hybrid models for shear resistance prediction of grouped stud connectors in concrete using improved metaheuristic optimization techniques

Shear connectors are essential in ensuring a composite action in steel–concrete structures. An accurate estimation of the shear resistance of group stud connectors (GSC) is crucial for the design of steel–concrete composite structures. Overfitting problems arise when the artificial neural network (ANN) is trained using traditional optimization algorithms. However, metaheuristic algorithms maintain the equilibrium between exploitation and exploration in solving this problem. This study combines metaheuristic optimization algorithms, including classical particle swamp optimizer (PSO) and novel improved eliminate particle swamp optimizer (IEPSO) with ANN. The proposed algorithms was used to simulate the shear resistance of GSC. A total of 232 data points of push-out test results were collected. The results show that the ANN-IEPSO with simpler architectures outperforms ANN-PSO and traditional optimization algorithms. The efficacy of the metaheuristic algorithms was benchmarked against four existing design codes, including AASHTO, EC 4, JSCE, and GB50017. The global sensitivity analysis visualizes the relationship between the six predictors and target shear resistance and shows the interaction between the two predictors. The number of studs, stud diameter, and stud spacing are the most critical parameters in predicting the shear resistance of the GSC.