Estimating Axial Bearing Capacity of Driven Piles Using Tuned Random Forest Frameworks

In the process of designing pile foundations, it is essential to take the axial bearing capacity ( \({B}_{c}\) $B_c$ ) of the pile into consideration., where determination of this target requires extreme fields and experimental efforts along with its cost. The primary objective of this study was to investigate the possibility of using tree-based approaches in order to estimate the axial \({B}_{c}\) $B_c$ of piles. The goal of building Random Forests ( \(RF\) $\mathit{RF}$ ) models is to produce a strong and adaptable machine learning method that is capable of reliably and accurately completing tasks related to classification as well as regression. Enhanced precision in predicting feature significance, scalability, adaptability, and managing missing data are the main objectives of employing \(RF\) $\mathit{RF}$ . The accuracy of this model is very dependent on its hyperparameters, which are linked to the Coati optimizer ( \(CO\) $\mathit{CO}$ ) and Giant trevally optimizer ( \(GTO\) $\mathit{GTO}$ ) procedures (also called \(RF-C\) $RF-C$ and \(RF-G\) $RF-G$ ) in order to find the optimal combinations. In a database, there were 472 driven pile static load test results collected from previous papers. Specifically, the construction, validation, and testing phases of the proposed framework were carried out using the learning set (70%), validation set (15%), and evaluating set (15%) of the dataset. Moreover, the feature importance analysis is designed to assess the impact of each input variable on the axial \({B}_{c}\) $B_c$ of piles. \(RF-C\) $RF-C$ and \(RF-G\) $RF-G$ offer promising \({B}_{c}\) $B_c$ forecasting capabilities, where the \(RF-C\) $RF-C$ approach outperformed the \(RF-G\) $RF-G$ method in \({R}^{2}\) $R^2$ values, with values of 0.9876, 0.9781, and 0.9873.