Hybrid machine learning algorithms for estimating shear strength of steel-reinforced concrete composite shear walls

This study investigates the shear strength prediction of Steel Reinforced Concrete Composite Shear Walls (SRCCSWs) using hybrid machine learning models. SRCCSWs offer enhanced seismic performance, energy dissipation, resistance capacity, and ductility, but their application is limited due to incomplete design standards and insufficient research. This research addresses these gaps by employing Artificial Neural Networks (ANNs) optimized with four metaheuristic algorithms: Runge-Kutta Optimizer, Comprehensive Learning Particle Swarm Optimizer, Fick's Law Algorithm, and Genetic Algorithm. A dataset of 149 SRCCSW samples was utilized to develop and validate these models. Key input parameters, including wall dimensions, material strengths, and reinforcement ratios, were considered. The optimal model was identified and further analyzed using SHapley Additive exPlanations (SHAP) and sensitivity analysis to elucidate the relationship between input variables and shear strength predictions. The findings demonstrate that the hybrid ANN optimized by Fick's Law Algorithm significantly improves predictive accuracy compared to traditional empirical equations. This model achieved a mean absolute error (MAE) of 0.0982, root mean squared error (RMSE) of 0.1447, and coefficient of determination (R2) of 0.856, thus offering a robust tool for the design and analysis of SRCCSWs. Additionally, SHAP analysis identified wall length, wall thickness, concrete compressive strength, and axial load ratio as the most influential parameters affecting shear strength prediction. This research contributes to the field by providing a more accurate and reliable method for predicting SRCCSW shear strength, potentially facilitating wider adoption of these structural elements in seismic-resistant construction.