Optimizing Pile Bearing Capacity Prediction Using Specific Random Forest Models optimized by Meta-Heuristic Algorithms for Enhanced Geomechanically Applications

To achieve highly accurate predictions of Pile Bearing Capacity (PBC), the study employs a cutting-edge approach featuring Specific Random Forest (RF) prediction models, strategically enhanced with two potent meta-heuristic algorithms: the Snake Optimizer (SO) and the Equilibrium Optimizer (EO). The effective incorporation of meta-heuristic algorithms establishes a strong basis for significantly enhancing the accuracy and effectiveness of PBC estimation. To validate the effectiveness of this model, a comprehensive analysis is conducted, leveraging PBC samples gathered from diverse soil types derived from previously conducted stabilization tests. The results of this research unveil three distinct models: RFEO, RFSO, and an individual RF model. Each of these models imparts invaluable insights, enhancing the accuracy of PBC predictions. This study not only presents an efficient and time-saving methodology but also holds significant implications for various geomechanical applications, marking a notable advancement in PBC prediction techniques. The input variables of this study can be defined as Average Cohesion, Average Friction Angle, Average Soil Specific Weight, Average Pile-Soil Friction Angle, Flap Number, Pile Area, and Pile Length. The synergistic combination of specific RF models with meta-heuristic algorithms yields auspicious outcomes, paving the way for real-time PBC estimation across a broad spectrum of geological scenarios. Remarkably, the RFSO model exhibits exceptional performance, achieving an R2 value of 0.998 for the entire dataset while boasting the lowest RMSE of 109.43. Compared to the basic RF and RFEO models, the RFSO model consistently demonstrates superior predictive and generalization capabilities.