Data-Driven Modeling of Lateral and Cracking Loads in Confined Masonry Walls Using Machine Learning

Confined masonry (CM) is becoming a widely adopted construction building method even in earthquake-prone regions due to its economic viability, construction simplicity, and material availability. However, existing empirical models for predicting lateral and cracking loads often fall short due to varied material properties, detailing of confining elements and construction practices. In this study, machine learning (ML) algorithms, such as Extreme Gradient Boosting (XGB), Random Forest (RF), and Extremely Randomized Tree (ERT), were employed to predict the seismic performance of CM walls, focusing on maximum lateral load capacity and cracking load based on an experimental dataset from 84 published studies, with 59 samples for training and 25 for testing. Different material, load, geometrical, and reinforcement detailing, related to the lateral load capacity of CM, were considered. This study also compares the performance of the existing empirical equations against the proposed ML models. The ML models demonstrated strong predictive capabilities, outperforming empirical equations in both maximum lateral load and cracking load predictions, with XGBoost yielding the highest accuracy, reflected by R2 values of 0.903 for lateral load and 0.876 for cracking load predictions, and lowest the RMSE (28.742 for lateral and 23.982 for cracking load). Additionally, a comparative analysis shows that while some empirical equations produce reasonably accurate predictions, most exhibit significant deviations from experimental results. This study finally employs Partial Dependence Plot (PDP) analysis to explain the importance and contribution of the factors that influence the lateral strength, and concludes that ML models, especially XGBoost, are highly effective in capturing the complex behavior of CM walls under vertical and lateral loads, making them valuable tools for enhancing the accuracy of seismic performance evaluations.