Implementing ensemble learning methods to predict the shear strength of RC deep beams with/without web reinforcements

This paper presents a practical yet comprehensive implementation of the ensemble methods for prediction of the shear strength for reinforced concrete deep beams with/without web reinforcements. The fundamentals of the background of the ensemble machine learning methods are firstly introduced, and four typical ensemble machine learnning models such as random forest, adoptive boosting, gradient boosting regression tree and extreme gradient boosting are utlized in this study to obtain the predictive model. Then the implementation procedure using these methods to train a predictive model is given in details. The input data is split into training and testing sets, the 10-fold cross validation is used to evaluate the model performance, the grid search method is used to find the hyper-parameters, and the feature importance and partial dependence analysis are adopted as the interpretation of the model outputs. To use the ensemble methods to predict the shear strength of reinforced concrete deep beams, in total 271 test data was collected for training the models. The models all achieve good capacity in predicting the shear strength, and demonstrate superior performance over traditional machine learnning methods. Meanwhile, the classical mechanics-driven shear models are also employed as comparisons. The sensitivity of the key factors in ensemble models is analyzed and the importances of the input variables are obtained. It is shown that the ensemble machine learnning models are significantly superior to mechanics-driven models in both predicting accuracy and discrepancy.

Automated summary

This paper introduces a practical and comprehensive implementation of ensemble methods for predicting the shear strength of deep beams with or without web reinforcements. The study utilizes four typical ensemble machine learning models and demonstrates their superior performance over traditional machine learning methods in predicting accuracy and discrepancy.