Metaheuristic-based prediction of shear resistance of headed stud connectors embedded in concrete coupled with SHAP explainability

• The shear strength stud connector was predicted using diverse Input Features. • RF and LSTM models were optimized by algorithms such as PSO, MFOA, and WOA. • RF models constantly outperformed LSTM models in terms of predictive accuracy and robustness. • SHAP analysis revealed that A w , f cm , A s , and d s were the most influential features for shear strength predictions. Accurate prediction of the shear strength (Vu) of headed stud connectors embedded in concrete is critical for ensuring the safety and reliability of structural systems. The present study employed machine learning (ML) techniques, specifically Random Forest (RF) and Long Short-Term Memory (LSTM) models, enhanced by metaheuristic optimization algorithms such as Particle Swarm Optimization (PSO), Moth Flame Optimization Algorithm (MFOA), and Whale Optimization Algorithm (WOA), to address the complexities of predicting shear strength. Comprehensive datasets (1334) incorporating key input parameters related to geometric, material, and mechanical properties were sourced from the literature. SHapley Additive exPlanations (SHAP) analysis was employed to enhance the interpretability of the models by identifying the global and local importance of features. Moreover, five (5) evaluation matrices, including the coefficient of determination (R 2 ), the root mean square error (RMSE), the mean absolute error (MAE), the mean absolute percentage error (MAPE), mean absolute error (RMSE). Pearson correlation coefficient (PCC) was used to check the model's performance. The results demonstrate that RF models consistently outperformed LSTM models in predictive accuracy and generalizability. Among the optimized models, RF-WOA-C1 achieved the best performance for Combination 1 (C1), RF-WOA-C2 excelled in Combination 2 (C2), and RF-MFOA-C3 was the top performer for Combination 3 (C3). For Combination 4 (C4), the standalone RF-C4 model delivered exceptional results without requiring additional optimization. SHAP analysis revealed that the projected weld collar (Aw) area, compressive strength of concrete (cm), and stud parameters were the most influential features in predicting shear strength. The research provides significant information regarding the applicability of ML techniques for predicting shear strength. The research can save time and avoid the need for expensive experimental testing. The outcomes enlighten future research and serve as a guide for engineering practice. The findings emphasize the importance of prioritizing RF-based approaches, optimizing critical parameters, and employing SHAP for transparency.