Ensemble Extreme Gradient Boosting based models to predict the bearing capacity of micropile group

Abstract

In most cases in which non-allowable settlement or losing of bearing capacity has been encountered in geotechnical engineering, employing micropile usually leads to satisfactory outcomes. Considering the costly and time consuming of utilizing the experimental tests in order to determine the bearing capacity of micropile-cap footings, utilizing artificial intelligence based approaches is a reasonable process in this regard. Nevertheless, because the q/cu is almost complex, forecasting phenomena is challenable. In this regard, powerful approaches were developed employing the hybrid Extreme Gradient Boosting (XGBoost) model on the database to precise estimation of the output. For this purpose, 780 rows of data has been collected in order to construct a powerful dataset from various existing valid literatures. Besides, four recently published optimizing approaches containing Black Widow Optimization Algorithm (BWOA), Artificial Hummingbird Algorithm (AHA), Arithmetic Optimization Algorithm (AOA) and Fire Hawk Optimization (FHO) have been employed to reach the high fitness of internal parameters of the XGBoost model. By simulating XGBoost optimized models, it was revealed that the developed AHA−XGB model can be more accurate than that of the other models in predicting the q/cu. Scatter index, sensitivity and uncertainty analysis were conducted to identified the importance of various input parameters was created.