Diagonal failure analysis of unreinforced solid clay brick masonry walls: comparative analytical and statistical strength evaluations

• A new database of 116 diagonal compression tests on clay brick URM wall is compiled. • Statistical evaluation reveals systematic links between the strength and experimental factors. • Existing analytical models show significant bias and prediction scatter. • Seven machine learning models are developed to predict diagonal tension strength. • The random forest model achieves superior accuracy and lowest prediction scatter. Unreinforced masonry (URM) components constructed with solid clay bricks are intrinsically vulnerable to failure under seismic loading, despite being prevalent in many earthquake-prone countries. When box behaviour is ensured through e.g. adequate wall-to-diaphragm connections, in-plane (IP) shear failure often governs the building-scale seismic performance, which can be sudden and brittle. Accurate estimation of diagonal tension strength is therefore critical for evaluating failure mechanisms of URM buildings, assessing their seismic capacity, and informing effective retrofit strategies. To this end, this study presents the first comprehensive and critical review of experimental investigations on the diagonal tension strength of URM walls made of solid clay bricks, based on which a novel database of 116 experimental tests is compiled. Statistical evaluations across diverse configurations reveal that certain experimental factors, such as specimen origin, number of leaves, and wall size, exhibit clear trends in their influence on diagonal tension strength. Existing predictive equations are also assessed against the compiled data, where substantial conservative bias and significant prediction scatter are observed. To explore complementary approaches, seven machine learning algorithms are developed and compared, with hyperparameters optimized through Bayesian optimization. All data and models are made openly accessible to support future research and practical implementation. The outcomes of this study offer a robust benchmark dataset, a critical assessment of influencing factors and existing predictive models, and a demonstration of the potential of machine learning as a complementary tool for improving predictive capabilities in the seismic assessment of masonry structures.