Machine learning-based model for moment capacity prediction and reliability analysis of PSC beams

Abstract

This study aims to develop an accurate and reliable Natural Gradient Boosting (NGB) model to predict the moment capacity of the prestressed concrete (PSC) beams. For this purpose, a database of 188 experimental results is first collected from the tests in the literature. The base learner of the NGB model is optimized using the Jaya Algorithm (JA), Moth Flame Optimizer (MFO), Sine Cosine Algorithm (SCA), and Sparrow Search Algorithm (SSA). Accordingly, the optimal base learner is implemented in the NGB model to get the final results. The performance of the NGB model is compared with those of the recently popular and efficient ML algorithms, such as Random Forest (RF), Adaptive Gradient Boosting (AGB), Gradient Boosting (GB), and eXtreme Gradient Boosting (XGB) models and the design approach. The comparative results show that the NGB models with the SSA-DT base learner (R2 = 0.981, A10 = 0.897, RMSE = 5.040 kN.m, MAE = 4.132 kN.m in the test phase) outperform others. A resistance reduction factor (ϕ) value of 0.8 to 0.925 is recommended for the developed NGB model to achieve a specific reliability index (βT) value of 2.5 to 3.5 via reliability analysis. Finally, a web application is developed to facilitate the moment capacity prediction without time consumption and cost.