Robustness of hybrid light gradient boosting for concrete creep compliance prediction

Abstract

Concrete creep is one of the most crucial factors in concrete. A reliable prediction of concrete creep is vital for safe concrete structure design and maintenance. However, the theoretical and empirical models are convoluted and unreliable due to the complex time-dependent behavior of concrete creep. This study collects a comprehensive experimental database from the literature to develop a hybrid machine learning model that combines grey wolf optimizer (GWO) and light gradient boosting (LGB), namely GWO-LGB, for predicting precisely concrete creep compliance (J_creep). Three widely used empirical models and six baseline ensemble machine learning models are adopted to evaluate the efficacy of the developed hybrid GWO-LGB mode. The comparative results reveal that the hybrid GWO-LGB model produces more accuracy in predicting the J_creep than other models. In addition, the Shapley Additive exPlanations (SHAP) method is used to investigate the influence of input parameters on the J_creep. Finally, a web tool is created to apply the hybrid GWO-LGB model readily to predict the J_creep with new input data without cumbersome programming.