a machine learning framework for predicting maximum displacement of reinforced masonry shear walls under lateral loading

Accurate estimation of the maximum displacement capacity of masonry shear walls under lateral loading is essential for performance-based seismic design, yet conventional analytical and numerical approaches remain computationally intensive, sensitive to modeling assumptions, and highly dependent on expert interpretation. these limitations restrict their applicability for rapid assessment and design optimization. to address this challenge, this study proposes a machine learning (ml) framework that integrates predictive accuracy, interpretability, and mechanical validation. a database of 93 fully grouted masonry walls tested under cyclic displacement-controlled loading is utilized to develop a systematically optimized multi-layer perceptron artificial neural network (mlp-ann). the model incorporates geometric, reinforcement, material, and axial-load parameters under the assumption of rectangular, fully grouted walls with consistent boundary conditions. extensive architectural trials yielded an optimized ann achieving r² values of 0.98, 0.97, and 0.90 for training, validation, and testing datasets, respectively. complementary random forest (rf) analysis identified wall length, height, reinforcement ratios, masonry strength, and axial-load ratio as the most influential predictors governing displacement response. to verify the mechanical plausibility of the ml predictions, a finite element model (fem) of a representative specimen was developed, reproducing experimental backbone curves within 5–10% deviation. the combined ann–rf–fem framework offers a fast, interpretable, and reliable tool for evaluating seismic displacement capacity of masonry walls. future research should expand the dataset to include diverse wall geometries, boundary conditions, and materials, and explore hybrid ml–fem or physics-informed models to further improve generalization and design applicability.