Comparative analysis of machine learning models and standard codes for predicting compressive strength in hollow block masonry

Abstract This study employs machine learning models to predict the compressive strength of hollow block masonry and examines the key factors that influence this strength. Accurate prediction of its compressive strength is crucial for ensuring the safety of structures. The study expands the dataset to include all relevant factors affecting (compressive strength of masonry units and mortar, mortar thickness, prism height-to-thickness ratio, length-to-thickness ratio, net-to-gross area ratio of hollow blocks, and bedding type). It compares the prediction accuracy of existing standard code equations and published empirical expressions with machine learning models. The findings demonstrate that machine learning models, particularly XGBoost (XGB), outperform standard code equations and other models in predicting compressive strength. The XGB model achieves high accuracy (R² >0.92) and low error (RMSE < 2.14 MPa) for both training and validation datasets. Feature importance analysis reveals that the strength of the masonry units is the most dominant factor affecting compressive strength, followed by mortar strength and the prism’s height-to-thickness ratio. Mortar thickness, length-to-thickness ratio, net-to-gross area ratio, and bedding type have a lesser influence on the overall strength. This study highlights the potential of XGB for accurately and efficiently predicting the compressive strength of hollow block masonry. Engineers can optimize the material selection, wall thickness, and design for improved structural performance by focusing on the most critical factors identified through feature importance analysis.